JPH0793000A - Speech encoding device - Google Patents

Abstract

PURPOSE:To obtain the speech encoding device which can be improved in total sound source encoding characteristics by securing the possibility that a representative sound source, which is good in encoding characteristics and overlooked when a single representative sound source is used, from a frame. CONSTITUTION:When a pitch cycle 14 has a value, i.e., when a current processed frame is a voiced sound, a voiced sound source encoding means 8 extracts plural representative sound sources 16 by a multiple representative sound extracting means 9 according to previously set extraction standards. Then a sound encoding distortion calculating means 10 compares each of the representative sound sources 16, extracted by the multiple representative sound source extracting means 9, with respective code words used for specific encoding and finds and outputs distortion 17 in specific encoding between two, e.g. the least squared distance on a time waveform as to all combinations. Further, a sound source code selecting means 11 selects the smallest distortion among them and outputs a sound source code 19 corresponding to the code word in the case of it.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice coding device used for digital transmission or storage of voice signals.

[0002]

2. Description of the Related Art A voice signal converted into a digital signal at a predetermined sampling period is analyzed in frame units of a predetermined length and separated into spectrum shape information and a sound source signal, and when the voice signal is a voiced sound, A speech code that improves the coding efficiency of a voiced sound portion by expressing a sound source signal in a frame as a repetition of a typical one-pitch cycle length section by utilizing the appearance of similar waveforms in a pitch cycle The conversion method is disclosed in, for example, Japanese Patent Application Laid-Open No. 2-84699.

FIG. 6 is a block diagram showing the configuration of a conventional speech coding apparatus disclosed in Japanese Patent Laid-Open No. 2-84699.

As shown in FIG. 6, the speech coding apparatus includes a spectrum shape analyzing means 2 for analyzing a spectrum shape of a speech signal 1 which is digitalized at a predetermined sampling period in frame units of a predetermined length. It has pitch period extraction means 3 for obtaining the pitch period of the voice signal 1 and voiced / unvoiced voice determination means 4 for determining whether the voice signal is voiced or unvoiced. The spectrum shape analysis means 2 is connected to the spectrum shape information coding means 5 for coding the spectrum shape information 12, and the spectrum shape analysis means 2 is connected.
A voiced sound source coding means 8 is connected to the pitch period extraction means 3. The voiced sound source encoding means 8 uses the voice signal 1
Is a voiced sound, a representative sound source extraction unit 30 that extracts a representative sound source 32 having a one-pitch period length from the sound source signal 13 in or near the current frame from the pitch period 14 obtained by the pitch period extraction unit 3, The representative excitation 32 is encoded by a predetermined encoding operation and outputs an excitation code 19, and is composed of an excitation encoding means 31.

The pitch cycle extraction means 3 is also connected to a pitch cycle coding means 6 for coding the pitch cycle 14 obtained by the pitch cycle extraction means 3. Further, the voiced unvoiced determination means 4 is connected to the pitch period extraction means 3 and the voiced unvoiced information encoding means 7 for encoding the voiced unvoiced information 15 output from the voiced unvoiced determination means 4.

Next, the operation will be described.

The spectrum shape analyzing means 2 analyzes the input voice signal 1 of the current frame to obtain spectrum shape information 12, and calculates a sound source signal 13 from the obtained spectrum shape information 12 and the voice signal 1. The spectrum shape information coding means 5 then uses the spectrum shape information 12
Is encoded and the obtained spectrum shape code 18 is output.

In addition, the voiced / unvoiced determination means 4 uses the voice signal 1
Is analyzed to determine whether it is voiced sound or unvoiced sound, and the result is output as voiced unvoiced information 15. Then, the voiced unvoiced information encoding means 7 encodes this voiced unvoiced information 15 and outputs the obtained voiced unvoiced code 21.

Further, when the voiced unvoiced information 15 is voiced sound, the pitch period extraction means 3 performs a pitch period analysis on the voice signal 1 and outputs the obtained pitch period 14. Then, the pitch period encoding means 6 determines the pitch period 1
4 is encoded and the obtained pitch period code 20 is output.

On the other hand, when the pitch period 14 has a value, that is, when the voiced unvoiced information 15 is a voiced sound, the voiced sound source encoding means 8 first uses the representative sound source extraction means 30 to generate a sound source signal in the current frame. A section having a length of a pitch period 14 centered on the point where the amplitude is maximum is extracted from 13 as the representative sound source 32 and output. Then, the excitation encoding means 31 encodes the representative excitation 32 by a predetermined encoding operation and outputs the excitation code 19.

Here, the excitation codebook used in the excitation encoding means 31 is a typical one of the excitation signals of the one-pitch book extracted by the same procedure as the extraction procedure of the representative excitation 32. , The waveform of each codeword in the excitation codebook has its maximum amplitude point always located at the center of the waveform. In the encoding in the excitation encoding means 31, a distortion measure is set between the representative excitation 32 and each code word, and the distance, that is, the code at which the distortion at the time of encoding is minimized among all the code words. A word is selected and the sound source code 19, which is the index thereof, is output. The least square distance is used as the distortion measure.

[0012]

However, in the conventional speech encoding apparatus shown in FIG. 6, the extraction result of the representative excitation 32 obtained by the representative excitation extraction means 30 and the encoding characteristic of the excitation encoding means 31. However, it has a great influence on the sound quality of the final reproduced sound after decoding.

In the conventional speech encoding apparatus, the representative excitation extracting means 30 extracts one representative excitation 32 of a frame, and the representative excitation 32 is encoded by the excitation encoding apparatus 31. Here, the case where the representative excitation extracting means 30 extracts the excitation signal of the section of one pitch period length adjacent to the representative excitation 32 as the second representative excitation and the encoding is performed by the excitation encoding means 30. Think At the time of this encoding, it is conceivable that the coding distortion for the section obtained as the representative excitation 32 is large and the coding distortion for the second representative excitation is smaller than that of the representative excitation 32. Instead, it may be better to use the second representative excitation with smaller coding distortion, which has better coding characteristics. However, in the above-described conventional example, since only one representative sound source 32 is extracted in a frame, there is a problem that such a case is not taken into consideration.

The sound source signal 13 is a voiced sound,
Although it appears as a repetition of similar waveforms with a pitch period of 14, the waveform mode often changes suddenly, and in the representative sound source extraction means 30 that extracts only one representative sound source 32 from the sound source signal 13 in the frame, It may be difficult to extract the representative sound source 32 that sufficiently captures the condition of the sound source signal 13, and the sound source after decoding becomes unstable.

In addition, generally the digitized signal is
Even if the signals are digitized from the same analog signal at the same sampling period, the waveforms may be slightly different if the sampling points are displaced at intervals less than the sampling period.

FIG. 7 is a diagram for explaining how the waveform shifts due to the difference in the positions of the sampling points. This diagram shows a sampling in which a certain audio signal is sampled at a predetermined sampling cycle. Although the same sampling is performed as the subsequent audio signal a, the sampling point is slightly deviated, and the audio signal b after the sampling is performed.
This is a schematic representation of the situation.

The positions of the maximum amplitude points of the sampled audio signal a and the sampled audio signal b are slightly deviated. Here, when a part of both signals is extracted and compared with this maximum amplitude point as a reference, a deviation of less than the sampling period occurs between the waveforms of both signals.

FIG. 8 is a diagram for explaining how the above-mentioned waveform shift occurs over several sampling cycles. In this figure, a certain audio signal is sampled at a predetermined sampling cycle. The sampled audio signal c is the same as the sampled audio signal c, but the sampling point is slightly deviated from the sampled audio signal d.

The positions of the maximum amplitude points of the sampled audio signal c and the sampled audio signal d are
It is located at the left and right maximum points of the voice signal waveform before sampling, which shows a bimodal shape. Here, when a part of both signals is extracted and compared with this maximum amplitude point as a reference, a shift of two sampling periods occurs between the waveforms of both signals.

Now, considering the case where the same speech signal as when a certain codeword in the excitation codebook was created is input to the conventional speech coding apparatus, it is the same speech signal.
It is desirable to select this code word, but the distortion becomes large due to the effect of a slight shift in the sampling points when the code word is created and when the representative sound source 32 is extracted,
There is a problem that different codewords may be selected, and the coding characteristics may deteriorate due to the influence of the deviation of the sampling points.

The present invention has been made in order to solve the above-mentioned problems, and a representative sound source having a better coding characteristic than that which is missed when a single representative sound source is used can be extracted from within a frame. It is an object of the present invention to obtain a speech coder that secures the possibility and improves the overall excitation coding characteristic. Another object of the present invention is to obtain a speech coding apparatus capable of improving the stability of the sound source after decoding by reflecting the state of the average sound source signal waveform in the frame on the coding of the representative sound source. Further, another object of the present invention is to obtain a speech coding apparatus capable of absorbing a small sample shift between the extracted representative sound source and the coded sound at the time of sound source coding and improving the coding characteristic of the sound source.

[0022]

According to a first aspect of the present invention, there is provided a speech coding apparatus, wherein a plurality of representative sound source extracting means for extracting a plurality of representative sound sources from a sound source signal and outputting the plurality of representative sound sources, and a plurality of representative sound source extracting means. Excitation coding distortion calculation means for calculating all distortions when each representative excitation is encoded into each excitation code that can be output by a predetermined encoding operation, and distortions calculated by the excitation coding distortion calculation means are compared, The present invention is characterized by further comprising excitation code selection means for selecting a combination of the representative excitation and the excitation code having the minimum distortion and outputting the excitation code.

According to a second aspect of the present invention, a speech coding apparatus extracts a plurality of representative sound sources by extracting a plurality of representative sound sources from a sound source signal and outputs the representative sound sources extracted by the plurality of representative sound source extracting means. Excitation coded distortion calculation means for calculating all distortions when encoded into each excitation code that can be output by encoding operation, and excitation code by multiplying each distortion calculated by the excitation coded distortion calculation means by a predetermined weighting coefficient Excitation code determination means for taking the sum for each and outputting the excitation code with the minimum sum is provided.

A speech encoding apparatus according to a third aspect of the invention is a plurality of representative sound source extracting means for extracting a plurality of representative sound sources from a sound source signal and outputting the plurality of representative sound sources, and averaging of each representative sound source extracted by the plurality of representative sound source extracting means. An averaged representative sound source calculation means for outputting one averaged representative sound source obtained by processing, and a sound source code for performing a predetermined coding operation on the averaged representative sound source to minimize distortion during coding. And an averaging excitation coding means for performing the same.

According to a fourth aspect of the present invention, there is provided a speech coding apparatus, wherein a representative sound source extraction position is determined by the plural representative sound source extraction means by determining one representative sound source extraction position from a sound source signal in a current frame and outputting the representative sound source extraction position. Pitch for extracting a plurality of representative sound sources of one pitch period length from the means, the representative sound source at the representative sound source extraction position, and the sound source at a position separated from the representative sound source extraction position forward or backward by an integer multiple of the pitch period. It is characterized in that a periodic interval representative sound source extraction means is provided.

In the speech coder according to the fifth aspect of the present invention, the representative sound source extraction position determination means for determining one representative sound source extraction position from the sound source signals in the current frame and outputting it to the plural representative sound source extraction means. And a displacement representative sound source extraction means for extracting a plurality of representative sound sources of one pitch cycle length from the sound source signals at those positions by giving a plurality of kinds of front and rear movements that are minute compared to the pitch period to the representative sound source extraction position. And is provided.

[0027]

In the speech coding apparatus according to the first aspect of the present invention, the plurality of representative sound source extraction means extracts a plurality of representative sound sources from the sound source signal and outputs the same, and the sound source coding distortion calculation means extracts the plurality of representative sound source extraction means. Calculates all distortions when each representative excitation is encoded into each excitation code that can be output by a predetermined encoding operation, compares the distortions calculated by the excitation coding distortion calculation means by the excitation code selection means, and finds that distortion The smallest combination of the representative excitation and the excitation code is selected and the excitation code is output.

In the speech coder according to the second aspect of the present invention, a plurality of representative excitation extracting means extracts a plurality of representative excitations from the excitation signal and outputs the representative excitations, and an excitation coding distortion calculating means extracts a plurality of representative excitation extracting means. All distortions when each representative excitation is encoded into each excitation code that can be output by a predetermined encoding operation are calculated, and a predetermined weighting coefficient is assigned to each distortion calculated by the excitation coding distortion calculation means by the excitation code determination means. The sum is multiplied to obtain the sum for each excitation code, and the excitation code having the minimum sum is output.

In the speech coding apparatus according to the third aspect of the present invention, the plurality of representative sound source extracting means extracts a plurality of representative sound sources from the sound source signal and outputs the same, and the averaged representative sound source calculating means extracts the plurality of representative sound source extracting means. Output one averaged representative sound source obtained by averaging each representative sound source,
The averaging excitation encoding means performs a predetermined encoding operation on the averaged representative excitation to output an excitation code that minimizes distortion during encoding.

In the speech coding apparatus according to the present invention, the representative sound source extraction position determining means determines and outputs one representative sound source extraction position from the sound source signal in the current frame, and extracts the pitch period interval representative sound source. The means extracts a plurality of representative sound sources having a one-pitch cycle length from the representative sound source at the representative sound source extraction position and the sound source at a position separated from the representative sound source extraction position by an integral multiple of the pitch period in the forward direction or the backward direction.

In the speech coder according to the fifth aspect of the present invention, the representative excitation extraction position determining means determines one representative excitation extraction position from the excitation signal in the current frame and outputs the displacement representative excitation extraction means. A plurality of types of front and rear movements that are minute compared to the pitch period are given to the representative sound source extraction position, and a plurality of representative sound sources of one pitch cycle length are extracted from the sound source signals at those positions.

[0032]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

Example 1. FIG. 1 is a block diagram showing a configuration of a speech coder according to the invention described in claim 1.

As shown in FIG. 1, the speech coding apparatus includes a spectrum shape analyzing means 2 for analyzing a spectrum shape of a speech signal 1 which is digitalized at a predetermined sampling period in frame units of a predetermined length. It has pitch period extraction means 3 for obtaining the pitch period of the voice signal 1 and voiced / unvoiced voice determination means 4 for determining whether the voice signal is voiced or unvoiced. The spectrum shape analysis means 2 is connected to the spectrum shape information coding means 5 for coding the spectrum shape information 12, and the spectrum shape analysis means 2 is connected.
A voiced sound source coding means 8 is connected to the pitch period extraction means 3. The voiced sound source encoding means 8 uses the sound source signal 1
3, a plurality of representative sound source extraction means 9 for extracting and outputting a plurality of representative sound source 16 which are a plurality of representative sound sources, and a plurality of representative sound source 16 extracted by the plurality of representative sound source extraction means 9 can be output by a predetermined encoding operation. Excitation coded distortion calculation means 10 for calculating all distortions when coded into the excitation code and distortion 17 calculated by the excitation coded distortion calculation means 10 are compared, and the representative excitation and the excitation code with the minimum distortion 17 are compared. And a sound source code selecting means 11 for selecting a combination of and outputting the sound source code 19.

Further, the pitch cycle extraction means 3 is connected to a pitch cycle coding means 6 for coding the pitch cycle 14 obtained by the pitch cycle extraction means 3. Further, the voiced unvoiced determination means 4 is connected to the pitch period extraction means 3 and the voiced unvoiced information encoding means 7 for encoding the voiced unvoiced information 15 output from the voiced unvoiced determination means 4.

Next, the operation of this embodiment will be described.

The spectrum shape analyzing means 2 analyzes the input voice signal 1 of the current frame to obtain spectrum shape information 12, and calculates a sound source signal 13 from the obtained spectrum shape information 12 and the voice signal 1. The spectrum shape information coding means 5 then uses the spectrum shape information 12
Is encoded and the obtained spectrum shape code 18 is output.

In addition, the voiced / unvoiced determination means 4 uses the voice signal 1
Is analyzed to determine whether it is voiced sound or unvoiced sound, and the result is output as voiced unvoiced information 15. Then, the voiced unvoiced information encoding means 7 encodes this voiced unvoiced information 15 and outputs the obtained voiced unvoiced code 21.

Further, the pitch period extracting means 3 performs pitch period analysis on the voice signal 1 when the voiced unvoiced information 15 is voiced sound, and outputs the obtained pitch period 14. Then, the pitch period encoding means 6 determines the pitch period 1
4 is encoded and the obtained pitch period code 20 is output.

On the other hand, when the pitch period 14 has a value, that is, when the current processing frame is a voiced sound, the voiced sound source encoding means 8 causes the plural representative sound source extraction means 9 to select a plurality of representative sound sources in accordance with preset extraction criteria. The sound source 16 is extracted. Then, the excitation coding distortion calculation means 10 compares each of the plural representative excitations 16 extracted by the plural representative excitation extraction means 9 with each codeword used for predetermined encoding,
The distortion 17 at the time of predetermined coding between the two, for example, the least square distance on the time waveform is obtained and output for all combinations. Further, the excitation code selection means 11 selects the minimum of these distortions 17 and outputs the excitation code 19 corresponding to the code word in that case.

Example 2. FIG. 2 is a block diagram showing the configuration of the speech encoding apparatus according to the invention of claim 2. It should be noted that the same components as those in FIG.

The voiced sound source encoding means 8 extracts a plurality of representative sound sources 16 from the sound source signal 13 and outputs them, and a plurality of representative sound sources 16 extracted by the plurality of representative sound source extraction means 9. Excitation coding distortion calculation means 10 for calculating all distortions when encoded into each excitation code that can be output by the encoding operation, and each distortion calculated by the excitation coding distortion calculation means 10 is multiplied by a predetermined weighting coefficient. And the excitation code determination means 22 for taking the sum for each excitation code and outputting the excitation code 19 with the minimum sum.

Next, the operation of this embodiment will be described.

The voiced sound source encoding means 8 has a pitch period of 14
Has a value, that is, when the current processing frame is a voiced sound, the plural representative sound source extracting means 9 extracts the plural representative sound sources 16 according to a preset extraction criterion. Then, the excitation coding distortion calculation means 10 compares each of the plurality of representative excitations 16 extracted by the plurality of representative excitation extraction means 9 with each codeword used for a predetermined encoding, and determines a predetermined value between the two. The distortion 17 at the time of encoding, for example, the least square distance on the time waveform is obtained and output for all combinations.
Further, the excitation code determination means 22 assigns a weighting factor to these distortions 17 according to the extraction position of each representative excitation (here, a large weighting factor is applied to the representative excitation extracted from a location near the center of the frame. ), The sum is calculated for each excitation code, and the excitation code 19 having the minimum sum is selected and output.

Example 3. In the above-described second embodiment, regarding the weighting coefficient assignment in the excitation code determination means 22,
A large weighting factor is given to the representative sound source extracted from a location close to the center of the frame, but not limited to this, the weighting factor of the weighting factor given according to the correlation between the previous frame representative sound source and each representative sound source is You may change the size.

Example 4. In the above-described second embodiment, regarding the weighting coefficient assignment in the excitation code determination means 22,
A large weighting coefficient is given to the representative sound source extracted from a location near the center of the frame, or the size of the weighting coefficient to be given is changed according to the correlation between the previous frame representative sound source and each representative sound source. The representative sound sources may be uniformly weighted to obtain the sum.

Example 5. FIG. 3 is a block diagram showing the configuration of a speech encoding apparatus according to the invention of claim 3. It should be noted that the same components as those in FIG.

The voiced sound source encoding means 8 performs a plurality of representative sound source extraction means 9 for extracting and outputting a plurality of representative sound sources 16 from the sound source signal 13, and an averaging process of each representative sound source extracted by the plurality of representative sound source extraction means 9. An averaged representative sound source calculation unit 23 that outputs one averaged representative sound source 25 obtained by performing the sound source code 19 that performs a predetermined coding operation on the averaged representative sound source 25 to minimize distortion at the time of coding. And averaging excitation coding means 24 for outputting.

Next, the operation of this embodiment will be described.

The voiced sound source encoding means 8 has a pitch period of 14
Has a value, that is, when the current processing frame is a voiced sound, the plural representative sound source extracting means 9 extracts the plural representative sound sources 16 according to a preset extraction criterion. Then, the averaged representative sound source calculation means 23 obtains the average value of the amplitude of each sample on the time waveform of each representative sound source extracted by the plural representative sound source extraction means 9, and calculates the averaged representative sound source 25.
Is calculated and output. Furthermore, averaging excitation coding means 24
Is the least square distance on the time waveform between the averaged representative excitation 25 and each code used for predetermined encoding,
The excitation code 19 corresponding to the code word that minimizes this is output.

Example 6. In the fifth embodiment described above, the averaged excitation coding means 24 does not bias the calculation of the average value, but the extraction position of each representative excitation and the correlation with the representative excitation of the previous frame are used as weighting criteria. The weighted average used may be obtained.

Example 7. In the first, second and fifth embodiments described above, the distortion measure of coding in the averaging excitation coding means 24 is the least square distance on the time waveform, but the distortion measure is not limited to this, and for example, Batacharia between waveforms. A configuration using distance or the like is also possible. Further, a configuration in which a distortion measure on the DFT spectrum is defined and distortion is calculated is also possible.

Example 8. In the first, second and fifth embodiments described above, the averaged excitation coding means 24 calculates the averaged representative excitation 25 by averaging the amplitudes on the time waveform. However, each of the plurality of representative excitations 16 is subjected to DFT and DFT. It is also possible to adopt a configuration in which the average is taken on the spectrum and the coding is performed using the distortion measure as in the seventh embodiment as it is.

Example 9. FIG. 4 is a block diagram showing the configuration of the plural representative excitation extracting means 9 of the speech coding apparatus according to the fourth aspect of the invention.

The plural representative sound source extraction means 9 determines the representative sound source extraction position from the sound source signal in the current frame and outputs it, the representative sound source extraction position determining means 26, the representative sound source at the representative sound source extraction position and its representative. Pitch cycle interval representative sound source extraction means 27 for extracting a plurality of representative sound sources of one pitch cycle length from a sound source extraction position and a sound source located at a position approximately an integer multiple of the pitch cycle forward or backward. .

Next, the operation of this embodiment will be described.

When the pitch period 14 has a value, that is, when the current processing frame is a voiced sound, the representative sound source extraction position determining means 26 determines the position of the maximum amplitude peak point from the entire region of the sound source signal 13 in the frame, And a 1-pitch cycle section centered on that point is output as extraction position information 28. Then, the pitch period interval representative sound source extraction means 2
7 is centered on one pitch period section of the extraction position information 28, and the maximum amplitude point within ± 2 to 3 points of the point which is an integer multiple of the pitch period away from the center of the extraction position information 28 from the whole area in the frame before and after it. Extract one pitch period section,
It is output as a plurality of representative sound sources 16. At this time, in the extraction section near the start and end of the frame, it may be extracted over the preceding and following frames.

Example 10. FIG. 5 is a block diagram showing the configuration of the plural representative excitation extracting means 9 of the speech encoding apparatus according to the fifth aspect of the invention.

The plural representative sound source extraction means 9 determines a representative sound source extraction position from the sound source signal in the current frame and outputs the representative sound source extraction position determining means 26, and a pitch period with respect to the representative sound source extraction position. Compared with the sound source signals at those positions, a plurality of 1
Displacement representative sound source extraction means 29 for extracting a representative sound source having a pitch cycle length is provided.

Next, the operation of this embodiment will be described.

When the pitch period 14 has a value, that is, when the current processing frame is a voiced sound, the representative sound source extraction position determining means 26 determines the position of the maximum amplitude peak point from the entire region of the sound source signal 13 in the frame, And a 1-pitch cycle section centered on that point is output as extraction position information 28. Then, the displacement representative sound source extraction means 29 extracts the 1-pitch cycle section of the extracted position information 28 and the 1-pitch cycle section about ± 1 to 3 points as an independent section, and outputs it as a plurality of representative sound sources 16. At this time, in the extraction section near the start and end of the frame, it may be extracted over the preceding and following frames.

Example 11. In the above-mentioned Examples 9 and 10,
The extraction position information 28 is determined based on the maximum amplitude point in the entire area of the frame. The reference is a position where the cross-correlation value of one of the representative sound sources in the previous frame and the sound source signal 13 in the frame becomes maximum. It is also possible to

Example 12 In the tenth embodiment described above, the displacement of the extraction position is obtained by moving the extraction segment on the sound source signal 13. However, the DFT spectrum is obtained by performing DFT on the one-pitch period segment centered on the position obtained from the extraction position information 28. A plurality of types of phase rotations on the spectrum corresponding to displacements of ± 3 points or less in the time direction are generated stepwise in a pseudo manner in the spectrum, and each is output as a plurality of representative sound sources 16. A configuration is also possible. With this configuration, it is possible to artificially create a displacement shorter than the sampling period.

Example 13. In the tenth embodiment described above, a plurality of representative sound sources are obtained by changing the single extraction position, but it is also possible to combine the configuration according to claim 4.

[0065]

As described above, according to the invention of claim 1, a plurality of representative sound source extraction means extracts a plurality of representative sound sources from a sound source signal and outputs the same, and a sound source coding distortion calculation means makes a plurality of representative sound source. All distortions when each representative excitation extracted by the extraction means is encoded into each excitation code that can be output by a predetermined encoding operation are calculated, and the distortions calculated by the excitation coding distortion calculation means by the excitation code selection means are compared. However, since it is configured to output the excitation code by selecting the combination of the representative excitation and the excitation code that minimizes the distortion, the coding characteristics that are overlooked when encoding using a single representative excitation Can be used for encoding, and the encoding characteristics can be improved comprehensively.

According to the second aspect of the invention, a plurality of representative sound sources are extracted from the sound source signal by the plurality of representative sound source extracting means and output, and the representative sound source extracted by the plurality of representative sound source extracting means by the sound source coding distortion calculating means. Is calculated by encoding each excitation code that can be output by a predetermined encoding operation, and the excitation code determination means multiplies each distortion calculated by the excitation coding distortion calculation means by a predetermined weighting factor Take the sum for each sign,
Since the excitation code whose sum is the minimum is output, the state of the average excitation signal waveform in the frame is reflected in the encoding as compared with the case where the encoding is performed using a single representative excitation. Therefore, the stability of the sound source after decoding can be improved.

According to the third aspect of the present invention, a plurality of representative sound sources are extracted from the sound source signal by the plurality of representative sound source extracting means and output, and each representative sound source extracted by the plurality of representative sound source extracting means by the averaged representative sound source calculating means. A sound source that outputs one averaged representative sound source obtained by performing the averaging process and performs a predetermined coding operation on the averaged representative sound source by the averaging sound source coding means to minimize distortion at the time of coding. Since it is configured to output the code, the state of the average excitation signal waveform in the frame can be reflected in the encoding, compared with the case where the encoding is performed using a single representative excitation, The stability of the sound source can be improved.

According to the fourth aspect of the present invention, the representative sound source extraction position determining means determines and outputs one representative sound source extraction position from the sound source signal in the current frame, and the pitch period interval representative sound source extracting means represents the representative sound source extraction position. It is configured to extract a plurality of representative sound sources of one pitch cycle length from the representative sound source at the sound source extraction position and the sound source at a position that is distant from the representative sound source extraction position forward or backward by an integer multiple of the pitch cycle. Therefore, the positional relationship of the sampling point with respect to its amplitude peak among multiple representative sound sources is often the same as that after coding, against the influence of the deviation of the sampling points seen when using a single representative sound source as in the past. It is possible to use another representative sound source that matches and exhibits a similar waveform, and it is possible to improve the coding characteristic.

According to the fifth aspect of the invention, the representative sound source extraction position determining means determines and outputs one representative sound source extraction position from the sound source signal in the current frame, and the displacement representative sound source extracting means extracts the representative sound source. Since a plurality of types of front and rear movements that are minute compared to the pitch period are given to the position and a plurality of representative sound sources of one pitch period length are extracted from the sound source signals at those positions, the deviation of the sampling points A representative sound source that can avoid the influence can be extracted, and the coding characteristic can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a speech encoding apparatus according to the invention described in claim 1.

[Fig. 2] Fig. 2 is a block diagram showing a configuration of a speech coding apparatus according to a second aspect of the invention.

FIG. 3 is a block diagram showing a configuration of a speech encoding apparatus according to the invention of claim 3.

FIG. 4 is a block diagram showing a configuration of a speech encoding apparatus according to a fourth aspect of the invention.

[Fig. 5] Fig. 5 is a block diagram showing the structure of a speech encoding apparatus according to the invention of claim 5.

FIG. 6 is a block diagram showing a configuration of a conventional speech encoding device.

[Fig. 7] Fig. 7 is a diagram for explaining a mechanism in which a shift of less than a sampling interval of a waveform occurs in a conventional speech encoding device.

[Fig. 8] Fig. 8 is a diagram for explaining a mechanism in which a shift of about a sampling interval of a waveform occurs in a conventional speech encoding device.

[Explanation of symbols]

 2 spectrum shape analysis means 3 pitch period extraction means 4 voiced unvoiced determination means 5 spectrum shape information coding means 6 pitch period coding means 7 voiced unvoiced information coding means 8 voiced sound source coding means 9 multiple representative sound source extraction means 10 sound source codes Distortion distortion calculation means 11 Excitation code selection means 22 Excitation code determination means 23 Averaged representative excitation calculation means 24 Averaged excitation coding means 26 Representative excitation extraction position determination means 27 Pitch cycle interval representative excitation extraction means 29 Displacement representative excitation extraction means

Claims (5)

[Claims] 1. A voice signal converted into a digital signal at a predetermined sampling period is analyzed in units of frames of a predetermined length and separated into spectrum shape information and a sound source signal, and when the voice signal is a voiced sound, A pitch period is obtained, a representative excitation having a pitch length of 1 is extracted from the excitation signal in or near the current frame, the representative excitation is encoded by a predetermined encoding operation, and an excitation code is output. ,
In a speech coder for encoding and outputting a plurality of parameters including spectrum shape information and voiced / unvoiced information, a plurality of representative sound source extraction means for extracting and outputting a plurality of representative sound sources from a sound source signal and a plurality of representative sound source extraction means are extracted. Comparing the distortion calculated by the excitation coding distortion calculation means and the excitation coding distortion calculation means for calculating all the distortions when each representative excitation coded into each excitation code that can be output by the predetermined coding operation, A speech encoding apparatus, comprising: an excitation code selecting unit that selects a combination of a representative excitation and an excitation code having the minimum distortion and outputs the excitation code. 2. A voice signal converted into a digital signal at a predetermined sampling period is analyzed in frame units of a predetermined length and separated into spectrum shape information and a sound source signal, and when the voice signal is a voiced sound, A pitch period is obtained, a representative excitation having a pitch length of 1 is extracted from the excitation signal in or near the current frame, the representative excitation is encoded by a predetermined encoding operation, and an excitation code is output. ,
In a speech coder for encoding and outputting a plurality of parameters including spectrum shape information and voiced / unvoiced information, a plurality of representative sound source extraction means for extracting and outputting a plurality of representative sound sources from a sound source signal and a plurality of representative sound source extraction means are extracted. Excitation coding distortion calculation means for calculating all distortions when each representative excitation is coded into each excitation code that can be output by a predetermined encoding operation, and each distortion calculated by the excitation coding distortion calculation means is predetermined for each distortion. A speech coding apparatus comprising: a sound source code determining unit that multiplies a weighting coefficient to obtain a sum for each sound source code, and outputs a sound source code having the smallest sum. 3. A voice signal converted into a digital signal at a predetermined sampling period is analyzed in frame units of a predetermined length and separated into spectrum shape information and a sound source signal, and when the voice signal is a voiced sound, A pitch period is obtained, a representative excitation having a pitch length of 1 is extracted from the excitation signal in or near the current frame, the representative excitation is encoded by a predetermined encoding operation, and an excitation code is output. ,
In a speech coder for encoding and outputting a plurality of parameters including spectrum shape information and voiced / unvoiced information, a plurality of representative sound source extraction means for extracting and outputting a plurality of representative sound sources from a sound source signal and a plurality of representative sound source extraction means are extracted. The averaged representative sound source calculation means for outputting one averaged representative sound source obtained by performing the averaging process of each of the representative sound sources, and the predetermined representative sound source is subjected to a predetermined coding operation to reduce distortion at the time of coding. A speech coding apparatus, comprising: an averaging excitation coding means for outputting an excitation code to be minimized. 4. The representative sound source extraction position deciding means for deciding and outputting one representative sound source extraction position from the sound source signal in the current frame, the plural representative sound source extracting means, and the representative sound source at the representative sound source extraction position Pitch cycle interval representative sound source extraction means for extracting a plurality of representative sound sources of one pitch cycle length from a sound source at a position approximately an integer multiple of the pitch cycle away from the representative sound source extraction position in the forward direction or the backward direction. The speech coding apparatus according to any one of claims 1 to 3, which is characterized. 5. The representative sound source extraction position deciding means for deciding and outputting one representative sound source extraction position from a sound source signal in a current frame, and the plurality of representative sound source extracting means have a pitch cycle with respect to the representative sound source extraction position. A displacement representative sound source extraction means for applying a plurality of minute forward and backward movements to extract a plurality of representative sound sources having a one-pitch cycle length from the sound source signals at those positions. The speech coding apparatus according to any one of claims 1 to 3.