JPH02176797A - Speech synthesis system - Google Patents

Abstract

PURPOSE:To decrease the amount of data of auxiliary information at a storage part by adding code data indicating the terminal of a segment and code data indicating the frequency of repetition of segment reproducing operation successively to the end of each segment and storing them. CONSTITUTION:The code data indicating the end of the segment of a waveform such as a pitch waveform segment and the code data indicating the frequency of repetition of segment reproducing operation are added successively right before or after speech code data at the end of the segment of the waveform and they are stored in a storage part 1. When read code data is speech code data, a composition part 5 decodes the speech waveform data and when the data is code data indicating the end of the segment, a repetitive resending processing part 4 outputs a necessary decision processing and control signal for the repetitive reproduction of the segment of a repetitive reproduction processing part 4, and the composition part 5 repeats segment reproducing operation from the speech code data of the head of the segment is repeated as many times as the contents of the sign data. Consequently, the amount of data of the auxiliary information stored on the storage part 1 can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial Application Field> The present invention relates to a speech synthesis method that generates synthesized speech by repeatedly reproducing waveform segments.

<Prior art> In the stationary parts of speech, such as vowels, if the bit length of the speech waveform voice code data, which is approximately the same at a period corresponding to the pitch of the sound, is about 3 to 4 bits, the above-mentioned segment The number of repetitions can be encoded with the same bit length as the voice code data.

However, the number of voice code data included in one elemental piece may be several hundred. For example, in order not to repeatedly play a segment at the transition part of speech (the middle part between phonemes), it is assumed that one segment at the transition part contains a waveform of 3 pitches, and the sampling frequency is Suppose that the frequency is 8KHz. Then, the pitch frequency is 10
01 (the number of voice code data when z is 8000x3/
+00 = 24.0 pieces. Therefore, in this case, the bit length required to represent the number of speech code data included in one segment is 8 bits, which is more than twice the bit length of speech code data, which is 3 to 4 bits. Become.

In other words, when the number of fragments stored in the storage section increases,
The number of bits occupied by the information on the number of voice code data included in each segment increases accordingly, and as the amount of auxiliary information stored in the storage section increases, a pitch structure in which data is repeated is observed.

Conventionally, code data representing the waveform of a one-pitch waveform segment in a speech waveform or a similar speech waveform segment is stored in advance in a storage unit, and the waveform of this segment is represented. There is a speech synthesis method that aims to reduce the amount of coded data of a speech waveform stored in a storage unit by decoding coded data repeatedly a necessary number of times to reproduce a synthesized speech waveform.

In the above speech synthesis method, speech code data obtained by encoding the waveform of each segment is stored in the storage unit. At that time, as auxiliary information when generating a synthesized speech waveform based on the speech code data of the segment (hereinafter referred to as segment reproduction), the number of repetitions of the segment, the number of speech code data included in one segment, and Information such as whether it is the final segment or not is also stored in the storage unit.

<Problems to be Solved by the Invention> Normally, among the auxiliary information used when reproducing a segment, the number of repetitions of a segment is about 2 to 4, and can be encoded with a bit length of 2 bits. Therefore, there is a noise problem.

SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a speech synthesis method that can reduce the amount of auxiliary information stored in a storage section by not storing the number of speech code data of a segment.

<Means for Solving the Problems> In order to achieve the above object, the present invention includes a storage unit that stores speech code data obtained by encoding a speech waveform for each waveform segment such as a pitch waveform segment; , in a speech synthesis method having a synthesis unit that generates a synthesized speech waveform by repeatedly reading and decoding the speech code data stored for each segment in the storage unit, each segment is Directly or immediately after the voice code data at the end of the segment, code data indicating the end of the segment and code data indicating the number of repetitions of the segment playback operation are successively added and stored in the storage unit; Each piece of code data stored in the storage section is read out one by one, and a piece end determination means determines whether the read code data is code data indicating the end of the piece or speech code data. If the segment end determining means determines that the read code data is voice code data, the synthesizer decodes the voice code data into voice waveform data, while decoding the read code data into voice waveform data. If it is determined that the encoded data is the code data indicating the end of the segment, the "-" synthesis unit repeats the code data at the beginning of the same segment the number of times indicated by the content of the code data indicating the number of repetitions. It is characterized in that it returns and repeats the above-mentioned segment reproduction operation.

Further, in the speech synthesis method, when the segment is the final segment stored in the storage unit, the segment is added to the segment immediately before or after the end speech code data of the final segment. Code data indicating the end of the segment and code data indicating the final segment that changes to code data indicating the number of repetitions are successively added and stored in the storage unit, and the segment end discriminating means reads the segment end. When it is determined that the code data is the code data indicating the end of the segment, the code data read next is code data indicating the final segment. They are read out one by one. Then, the segment end determining means determines whether the read code data is code data indicating the end of the segment or audio code data. As a result, if it is determined that the read code data is voice code data, the voice code data is decoded into voice waveform data by the synthesis unit, while the read code data is If it is determined that the code data indicates the end of the segment, the synthesizing unit returns to the speech code data at the beginning of the same segment the number of times indicated by the content of the code data indicating the number of repetitions, and returns to the speech code data at the beginning of the segment. The playback operation is executed repeatedly.

Therefore, even without using code data indicating the number of speech code data included in an element, the elemental piece reproduction operation can be performed on an elemental piece basis.

Further, in the speech synthesis method of the present invention, when the segment is the final segment stored in the storage section, immediately before or after the voice code data at the end of the final segment,
The final segment determining means determines whether the code data indicates the end of the segment or the code data indicating the number of repetitions, and the final segment discriminating means determines whether the read code data indicates the final segment. If it is determined that the read code data is code data, the synthesis unit ends the segment reproduction operation, while if it is determined that the read code data is code data indicating the number of repetitions, the synthesis unit The present invention is characterized in that the segment reproduction operation is repeated by returning to the first speech code data of the same segment as many times as indicated by the content of the code data indicating the number of repetitions.

Effects> In the speech synthesis method of the present invention, the code data indicating the end of the segment and the segment are stored in the storage section immediately before or after the voice code data at the end of the segment of the waveform such as the pitch waveform segment. The code data indicating the number of repetitions of the segment playback operation in which the stored voice code data is sequentially read and decoded is sequentially added and stored in the storage unit.

When performing the segment reproduction operation, data and code data indicating the final segment that changes to the code data indicating the number of repetitions are successively added and stored in the storage section.

When performing the segment reproduction operation, the code data stored in the storage section is sequentially read out one by one. If the segment end determination means determines that the read code data is code data indicating the end of the segment, the next read code data indicates the final segment. The final segment determining means determines whether the code data is the code data or the code data indicating the end of the segment.

As a result, if it is determined that the read code data is code data indicating the final segment, the synthesis section ends the segment reproduction operation. On the other hand, if the read code data is determined to be the code data indicating the end of the segment, the synthesizing section repeats the start of the same segment the number of times indicated by the content of the code data indicating the number of repetitions. The voice code data is returned to and the above segment reproduction operation is repeated.

Therefore, even without using code data indicating the number of speech code data included in an I segment, the segment reproduction operation can be performed on a segment-by-segment basis. In addition, without securing a separate area on the storage unit for code data indicating the final segment, it is determined that the reproduced segment is the final segment, and the 1 segment reproduction operation is terminated. can do.

<Example> Hereinafter, the present invention will be explained in detail with reference to illustrated embodiments.

FIG. 1 is a block diagram of a speech synthesis device according to the present invention. ROM (read-only memory) as a storage unit
1 converts analog audio waveform data into D PCM (differential pulse coding) and ADPCM by an encoder (not shown).
The address counter 2 stores coded data such as audio coded data and auxiliary information obtained by encoding using a waveform encoding method such as (adaptive differential pulse encoding), and the address counter 2 indicates an address when accessing the ROMI. This address counter 2 is set every time voice code data is read from ROMI.
I+ Voice code data C is stored in RQMI for each segment.

Then, the last speech code data C(1,n) of each segment,
. . . After C(L, m), code data E which does not indicate the end of the segment and code data R which indicates the number of repetitions when repeating the segment reproduction are stored. If segment reproduction is not repeated, the content of code data R indicating the number of repetitions is set to "0". In addition, if this elemental piece is the final elemental piece (L) that is the last elemental element stored in ROMI, the final elemental element (L) is stored in the location where code data R representing the number of repetitions is stored.
) is stored. Here, code data E indicating the end of the segment, code data R indicating the number of repetitions,
Code data RE and voice code data C indicating one final segment (L) must be different code data from each other. That is, for example, if the bit length of each of the code data E, R, and RE is 4 bits, then each of the code data E, R, and
, RE can be represented by numbers from 0 to 15. Of these, 0-14 is assigned to the voice code data C and the code data R indicating the number of repetitions, and "bi" is added to the code data E indicating the end of the segment and the contents of the address counter itself.Address stack 3 stores the value of the address on the ROMI where the first voice code data in the segment is stored.This address is stored in the address counter 2.
Supplied from.

The repetitive reproduction processing section 4 performs discrimination processing and outputs control signals necessary for repeatedly performing fragment reproduction.

The decoding unit 5 decodes the input voice code data and outputs digital synthesized voice waveform data.

The D/A converter 6 performs D/A conversion on the input digital synthesized voice waveform data and outputs analog voice waveform data. The parameter stack 7 stores parameters necessary for decoding the audio code data at the beginning of a segment. The repetition number counter 8 stores the number of repetitions of the same elemental piece during elemental piece reproduction, and subtracts the stored repetition number each time elemental piece reproduction based on the same elemental piece is repeated.

FIG. 2 shows an example of the format of code data such as voice code data stored in ROMI. It is sufficient to allocate 15 to code data RE indicating the final elemental piece obtained by encoding by an encoder (not shown).

Here, a segment is a speech waveform segment with one pitch period or a similar speech waveform segment when repeated playback is performed, and an interval between two segments that are repeatedly played back when repeated playback is not performed. This is a speech waveform segment.

The speech synthesis apparatus having the above configuration and having the ROMI in which each code data is stored in the format shown in FIG. 2(a) operates according to the flowchart of the segment playback operation shown in FIG. 3. Hereinafter, the segment reproduction operation will be explained in detail according to this flowchart.

In step S1, when the elemental piece reproduction operation is started, each part is initialized as follows. That is, the contents of the address counter 2 and the address stack 3 are set to the address of the ROM 1 storing the first voice code data C(1,1) of the first segment (1). Further, the contents of the parameter stack 7 are set to be the same as the initialized parameters stored in the decoding unit 5. Furthermore, the content of the repetition number counter 8 is set to "0".

In step S2, the RO indicated by address counter 2
The encoded data stored at the address M+ is read out and output to the repetitive reproduction processing section 4 and the decoding section 5. Then, "bi" is added to the content a of the address counter 2, and the address is updated to the address for reading code data from the ROM 1 next time.

In step S3, the repetitive reproduction processing unit 4 determines whether the input code data is code data E indicating the end of a segment or voice code data C. the result,
In the case of code data E indicating the end of a segment, the process proceeds to step S6, and in the case of speech code data C, the process proceeds to step S4.

In step S4, since the coded data read from the ROM 1 in step S2 is voice coded data C, this voice coded data C is decoded by the decoding section 5 to obtain digital synthesized voice waveform data.

In step S5, it is determined whether the code data obtained in step S4 is code data RE indicating the final segment (L) or code data RE indicating the number of repetitions of reproduction. As a result, if it is determined that the code data RE indicates the final elemental piece (L), the elemental piece reproduction operation is ended; otherwise, the process advances to step S8.

In step S8, the content r of the repetition number counter 8 is “0”.
If the result is "θ", the process advances to step S9; otherwise, the process advances to step S12.

In step S9, the content of the code data R indicating the number of repetitions read in step S6 is set in the repetition number counter 8.

In step SIO, the content r of the repetition number counter 8 is θ
If the result is "0", the process proceeds to step Sll and preparations are made to reproduce the next elemental piece, while if not, the process proceeds to step S14 and the same elemental piece is reproduced. Preparation begins for repeated fragment reproduction based on fragments.

At step SLI, the digital synthesized speech waveform data in the next segment of the ROM 1 is D/A converted by the D/A converter 6, and an analog synthesized speech waveform is output. Then, the process returns to step S2 and begins processing the voice code data at the next address.

On the other hand, if it is determined in step S3 that the input code data is code data E representing the end of the segment, the following processes from step 86 to step S14 are executed by the repeat reproduction processing unit 4. .

In step S6, the next code data is read from ROM1. In this case, the read code data is the code data read from the address next to the code data E indicating the end of the segment, so the code data R indicating the number of repetitions
Alternatively, it is code data RE indicating the final elemental piece. Then, "bi" is added to the content a of the address counter 2.

Then, the content a of the address counter 2 is updated to the address of the first speech code data C in the next segment.

In step S7, the address storing the address of the audio code data C at the beginning of the reading in step S6 is checked.
Set counter content a to address stack 3. Further, in this case, the decoding unit 5 also holds parameters for decoding the first speech code data C of the next segment. Then, the values pi of the parameters held in the decoding unit 5 are set in the parameter stack 7, and the process returns to step S2.

In step S12, "1" is subtracted from the content r of the repetition number counter 8.

In step S13, the content r of the repetition number counter 8 is θ
If the result is "0", the process advances to step Sll to prepare for the reproduction of the next elemental piece; otherwise, the process advances to step S14 to reproduce the elemental element in the same elemental piece. Begins preparation for repeating one-sided playback.

In step 814, the contents a° of the address stack 3 storing the address on ROM 1 of the first speech code data C in the fragment that has just been decoded is the address.
The counter 2 is set, and the parameters pi of the decoding unit 5 are updated to the contents pi of the parameter stack 7.

By doing so, it becomes possible to decode again the speech code data C at the beginning of the segment for which decoding has just been completed, and preparations for repeating segment reproduction based on the same segment are completed. Then, the process returns to step S2, and repetition of elemental piece reproduction is performed based on the same elemental piece.

The fragment reproduction operation will be explained in more detail below.

FIG. 5(a) shows an example of the contents of the ROMI, and it is assumed that each code data C has the same number as in FIG. 2(a). FIG. 5(b) shows a synthesized speech waveform output from the D/A converter 6 according to the contents of the ROM 1 shown in FIG. 5(a).

When the fragment reproduction operation starts, step S1 shown in FIG.
Then, the content a of address register 2 is a ROM in which the first code data C (1, 1) of the first segment (1) is stored.
The address of I becomes "0". Further, the content r of the repetition number counter 8 also becomes "0". Then, step 82~
The process of step S5 is repeated.

While repeating steps 82 to S5 to generate the synthesized speech waveform of segment (2), if code data E indicating the end of segment (2) is detected in step S3, the process advances to step S6. At step S6, code data R2 (=2) indicating the number of reproduction repetitions is obtained. Then, since the value of R3 is not "θ", the value of R2 is set to "2" in the repetition number counter 8 in step S9, and then the process proceeds to step S14. Then, step S14
First, the first speech code data C(
The contents of the address stack 3, which is the address "X" of 2, 1), are set in the address counter 2 again.

Furthermore, the speech code data C (2゜l
) are transferred from the parameter stack 7. Then, the process returns to step S2, and the audio code data C(2,1) at the beginning of the segment (2) is read out, and the segment reproduction operation based on the segment (2) is performed again.

At the same time, the speech code data C of the first segment (1) is read from the ROMI and a synthesized speech waveform is generated.

In this way, the last speech code data C(
1, n), the code data E indicating the end of the segment is read out in step S2, and the code data E indicating the end of the segment is read out in step S3.
The process then proceeds to step S6.

When the code data R, (-"0") indicating the number of reproduction repetitions is read out in step S6, the content r of the repetition number counter 8 becomes "0", so that the process proceeds to step S7 and step S8. Step S9. The process advances to step 811 via step SIO. Then, the content a of address counter 2 is the first speech code data C (2) of the next segment (2).
, 1), and this value "x" is set in the address stack 3 in step Sll. Further, the parameters pl used when the decoding unit 5 decodes the first speech code data C of the segment (2) are sent from the decoding unit 5 as parameters pl.
It is transferred to stack 7 and is cented. Then, the process returns to step S2 and repeats the fragment reproducing operation step 82 to step S5 based on the next fragment (2).
2) When generating the synthesized speech waveform, step S3
When code data E indicating the end of elemental piece (2) is detected at step S6, the process proceeds to step S7 and then to step 8. Since the content r of the repetition number counter 8 is "2", the process advances to step S12, and in step S12, the content of the repetition number counter 8 becomes "bi". Therefore,
Since the content of the repetition number counter 8 is not "0", similarly to the above, in step S14, the RO when decoding the speech code data C(2,1) at the beginning of the segment (2) is
The address “x” of MI is set in the address counter 2 three times, and the parameters pi of the decoding unit 5 are set in the address counter 2 three times.
is set three times. Thereafter, the process proceeds to step S2, where the audio code data C(2,1) at the beginning of the segment (2) is read out, and the segment reproduction operation based on the 3-carbon segment (2) is performed.

Steps 82 to S5 are repeated until the 3 carbon pieces (
2) When generating the synthesized speech waveform, step S3
When code data E indicating the end of elemental piece (2) is detected at step S6, the process proceeds to step SI2 via step S7 and step S8, and the content r of the repetition number counter 8 becomes "0", and the element The repetition of the elemental piece reproduction operation based on piece (2) ends. Then, the process advances from step 913 to step Sll.

Then, the content a of the address register 2 becomes the address "y" of the ROMI that stores the first speech code data C (3, 1) of the next segment (3), and this value "y" ” is added to address stack 3 in step Sll.
is set to Further, parameters p1 used when the decoding unit 5 decodes the first speech code data C(3,1) of the segment (3) are set in the parameter stack 7. Then, return to step S2 and proceed to the next elemental piece (3)
The elemental piece reproduction operation based on the above is executed.

Steps 82 to S5 are repeated, and the previous fragment (2
), the synthesized speech waveform of segment (3) is generated. At this time, it is determined that the decoding of the code (L, m) indicating the number of reproduction repetitions in the segment (3) stored in the ROMI has been completed, and the segment reproduction operation is ended.

In this way, as a result of executing the segment playback operation based on the contents of the ROMI shown in FIG. 5(a), as shown in FIG. 5(b), a synthesized speech waveform based on segment (1) is generated. Subsequently, a synthesized speech waveform based on segment (2) continues three times, and a synthesized speech waveform based on segment (3) (synthesized speech waveform continues twice) is then output.

As described above, in the speech synthesis method of the present invention, when the speech code data C is stored in the ROM 1, immediately after the speech code data C at the end of each segment, the code data E and the code data indicating the end of the segment are stored. Code data R indicating the number of reproduction repetitions is added using the same bits. Immediately after the audio code data C at the end of the final segment, code data RE indicating the final segment, which changes to code data E indicating the end of the segment and code data R indicating the number of repetitions of reproduction, is added in the same bit. do. Then, when playing a segment, when the audio code data C is sequentially read out and decoded, and the code data E indicating the end of the segment is read out, data R3 is "1", so the segment The segment reproduction operation based on (3) is repeated once. Then, when the synthesized speech waveform based on the segment (3) is output twice, the content a of the address register 2 becomes the speech code data C(
This is the address "Z" of the ROMI storing the data L, 1), and this value "2" is set in the address stack 3 in step Sll. Furthermore, the decoding unit 5
is the first speech code data C(L, 1) of the final segment (L)
The parameters pi used when decoding are the parameters
Set on stack 7. Then, the process returns to step S2, and an elemental piece reproduction operation based on the next final elemental piece (L) is executed.

In this way, when the synthesized speech waveform of the final segment (L) is generated by repeating steps 82 to S5, when code data E indicating the end of the final segment (L) is detected in step S3, the process proceeds to step S6. Then, in step S6, code data RE indicating the final elemental piece is read out. Then, in step S7, the segment reproduction based on the same segment is repeated a number of times according to the content of the code data R indicating the number of repetitions to be read next to the last audio code data C of the final segment (L). On the other hand, if the code data following the code data E indicating the end of a segment is the code data RE indicating the final segment, the segment reproduction operation is ended. Therefore, it is possible to perform segment-by-fragment playback without using information indicating the number of speech code data included in l segment, and code data indicating the number of speech code data included in l segment can be reproduced. There is no need to store it in the storage unit. Therefore, the amount of auxiliary information in the storage section can be reduced.

In the flowchart of FIG. 3 of the above embodiment, the content a of the address register 2 is set to the address in step S14.
When updating the address of the first voice code data C of the segment stored in the stack 3, and then reading the voice code data C from this address in step S2 and repeating the segment playback, the following problem occurs. There is. That is, R
If the access time of OM1 (the time from when the ROMI address is determined to when the ROMI data is output) is long, it is necessary to wait for a while until the voice code data C is read out in step S2. Therefore,
In the following embodiment, such an operation is not required.

In this embodiment, the ROMI format is
As shown in Figure (b), the code data E indicating the end of the segment other than the final segment (L) and the code data R indicating the number of repetitions of reproduction are converted into the last audio code data C(+,n). ,...
Add immediately before. Then, the steps after step S8 in the flowchart of FIG. 3 are performed as shown in FIG.

That is, in step S29, the content of the code data R indicating the number of repetitions read in step S6 is set in the repetition number counter 8.

In step S30, the content r of the repetition number counter 8 is 0.
If the result is "0", the process advances to step S31 to prepare for the next elemental piece reproduction operation; otherwise, the process advances to step S37 to reproduce the same elemental element. D/A by the repeater 6 of fragment-based fragment reproduction
After conversion, an analog synthesized speech waveform is output. Then, the process returns to step S2 and begins processing the voice code data at the next address.

In step S35, the content r of the repetition number counter 8 is “
1″ is subtracted.

At step 836, the content r of the repetition number counter 8 is “
0". If the result is "0", the process advances to step S31 and preparations are made for the reproduction of the next elemental piece; if not, the process advances to step S37 and the reproduction is based on the same elemental piece. Begins preparation for repeating fragment playback.

In step S37, the last speech code data C in the current segment is read out, and the address stack 3 stores the address on ROM 1 of the first speech code data C in the segment for which the previous decoding has been completed. The contents a' are set in address counter 2.

In step 82B, the ROM is
The last audio coded data C read from 1 is decoded by the decoding section 5, and preparations are made for digital return.

In step S31, the last speech code data C in the current element is read out, and "bi" is added to the content a of the address counter 2. Then, the content a of the address counter 2 becomes the next element. It is updated to the address of the first voice code data C in the piece.

In step 832, the ROM in step S31 above is
The last voice code data C read out from 1 is decoded by the decoding section 5 to obtain digital synthesized voice waveform data.

In step S33, the contents a of the address counter storing the address on the ROMI of the first audio code data C in the next segment are set in the address stack 3. Also, the value p of parameters when decoding the first speech code data of the next segment held in the decoding unit 5
i is set in parameter stack 7.

In step 834, the digital synthesized speech waveform data obtained in step S32 is converted into D/A synthesized speech waveform data.

In step S39, the parameters pi of the decoding unit 5 are updated to the contents pi of the parameter stack 7.

By doing this, it becomes possible to decode again the speech code data C at the head of the segment for which the previous decoding has been completed, and preparations for repeating segment reproduction based on the same segment are completed.

In step S40, the digital synthesized speech waveform data obtained in step 838 is D/A converted by the D/A converter 6, and an analog synthesized speech waveform is output. Then, the process returns to step S2 and starts processing the voice code data C at the next address.

That is, in this embodiment, when repeating elemental piece reproduction, the address counter 2 is set in step S37.
After setting the content a° of address stack 3 to , decoding processing is performed in step 938, and step S3
9, the contents pi° of the parameter stack 7 are set to the parameters pi of the decoding unit 5, and step S40
After performing D/A conversion processing in step S2, voice code data C is obtained from the address r'tOM+ set in the address counter 2 in step S37.
I am trying to read out the . Therefore, there is sufficient time after the address counter 2 is set to the contents a'' of the address stack 3 until the audio code data C is read from the address indicated by the contents a of the address counter 2. , even if the access time to ROM1 is long, the operation of waiting for readout of code data in step S2 is unnecessary.

<Effects of the Invention> As is clear from the above, the speech synthesis method of the present invention has the following effects:
Immediately or immediately after the audio code data at the end of each segment, code data indicating the end of the segment and code data indicating the number of repetitions of the segment playback operation are successively added and stored in a storage unit; If the coded data read from the storage unit is audio coded data, the audio coded data is decoded, while if it is coded data indicating the end of the segment, playback operation is performed based on the same segment. As will be repeated, Fig. 1 is a block diagram of a speech synthesis device according to the present invention, and Fig. 2 (
a) and FIG. 2(b) are diagrams showing an example of the A-mat (of code data stored in the ROM) shown in FIG. 1, and FIG.
The figure is a flowchart of the segment playback operation when using a ROM in which encoded data is stored in the format of Figure 2(a), and Figure 4 is a flowchart of the segment reproduction operation when encoded data is stored in the format of Figure 2(b). When using a ROM with FIG. 6 is a diagram showing a synthesized speech waveform obtained by performing a segment playback operation according to the contents of the ROM shown in FIG.

1-ROM, 2...address counter,
3-Address stack, 4.Repetitive playback processing section, 5.Decoding section, 6.D/A
Converter, 7. Parameter stack, 8. Repeat number counter.

Patent applicant Sharp Co., Ltd. agent Patent attorney Aoyama Aoyama and one other person,
It is possible to perform an elemental piece reproduction operation on an elemental piece basis without using code data indicating the number of audio code data included in one elemental piece. Therefore, the amount of auxiliary information in the storage section can be reduced.

Furthermore, in the speech synthesis method of the present invention, when the segment is a final segment, code data indicating the end of the segment is added immediately before or after the speech code data at the end of the final segment; Code data indicating the final segment to be changed to the code data indicating the number of repetitions are successively added and stored in the storage unit, and the code data read from the storage unit does not indicate the end of the segment. If the next read code data is the code data indicating the final segment, the segment reproduction operation is terminated. Therefore, the code data indicating the final segment is stored in the storage section. Since there is no need to secure a separate area, the amount of data in the storage section can be further reduced.

[Brief explanation of the drawing]

Claims (2)

[Claims] (1) a storage unit that stores voice code data obtained by encoding a voice waveform for each waveform element such as a pitch waveform element;
In a speech synthesis method having a synthesis unit that generates a synthesized speech waveform by repeatedly reading and decoding the speech code data stored for each segment in the storage unit, immediately before or after the end speech code data, code data indicating the end of the segment and code data indicating the number of repetitions of the segment playback operation are successively added and stored in the storage unit; Each piece of code data stored in the segment is read out one by one, and the segment end discriminating means determines whether the read code data is code data indicating the end of the segment or audio code data, and When the segment end determination means determines that the read code data is voice code data, the synthesizer decodes the voice code data into voice waveform data, while decoding the read code data. If it is determined that the data is code data indicating the end of the segment, the synthesizing unit returns to the speech code data at the beginning of the same segment the number of times indicated by the content of the code data indicating the number of repetitions. A speech synthesis method characterized by repeating one-sided playback operations. (2) In the speech synthesis method according to claim 1, if the segment is the final segment stored in the storage unit,
Immediately or immediately after the audio code data at the end of the final segment, code data indicating the end of the segment and code data indicating the final segment that changes to code data indicating the number of repetitions are successively added. When the segment end determination means determines that the read code data is code data indicating the end of the segment, the next read code data is stored in the segment end discriminating unit. A final segment discriminating means determines whether the coded data indicates a segment or the number of repetitions, and the final segment discriminating means determines that the read coded data is coded data indicating a final segment. If it is determined that
The synthesizing section ends the segment reproduction operation, and if the read code data is determined to be the code data indicating the repetition number, the synthesis section causes the content of the code data indicating the repetition number. A speech synthesis method characterized in that the above-mentioned segment reproduction operation is repeated by returning to the first speech code data of the same segment a number of times indicated by .