JPH11231898A - Speech synthesizing device and its control method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate a more natural accent by using information on a word positioned before a word for combining whenever necessary, in determining an accent position at the time of combining the word. SOLUTION: An accent combining table 110 holds the position of an accent for combining a word, using a Chinese numeral retrieval key composed of at least one numeral and an auxiliary numeral retrieval key composed of a numeral. At the time of aurally synthesizing an expression composed of the combination of a numeral and an auxiliary numeral, a Chinese numeral retrieval key generation treating part 104 selects an expression maximaly agreeing to the numerical expression in the aurally synthesized expression, from the Chinese numeral retrieval keys registered with the accent combining table 110 for use as the Chinese numeral retrieval key. On the other hand, an auxiliary numeral retrieval key generation treating part 108 generates the auxiliary numeral retrieval key according to auxiliary numerals in the expression to be aurally synthesized. Also, an accent combining tale retrieval treating part 106 retrieves the accent combining table 110, using the generated Chinese numeral retrieval key and the auxiliary numeral retrieval key, thereby obtaining an accent position at the time of the combining.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a speech synthesizer and a control method thereof. In particular, the present invention relates to a speech synthesizer capable of appropriately performing accent connection when two words are connected, and a control method thereof.

[0002]

2. Description of the Related Art In general, a speech synthesizer considers accent connections between words in order to generate a natural synthesized speech. An accent connection between words is an accent connection that occurs between a noun and a case particle, a verb and an auxiliary verb, a number and a classifier, and the like. In particular, realizing an accent connection between a number and a classifier is important in generating a natural synthesized speech.

[0003] Accent connection between a numeral and a classifier by a conventional device of this kind is performed in the following procedure. (1) Convert a number into standard Chinese numeral notation (e.g., 12345 → 12,300,345). (2) The accent binding table is referred to using the classifier and the kanji numeral ("5" in the example of (1)) immediately before the classifier as a key,
Determine the position of the accent when a number and a classifier are concatenated.

[0004]

However, in the conventional method of determining the position of the accent only by the classifier and the Chinese numeral just before the classifier, as shown in the following example, there are cases where the accent cannot be accurately expressed. I will.

FIG. 5 is a diagram showing an accent combination table of a classifier "en" and a numeral. In the accent combination table of FIG. 5, the position of the accent when a number and a classifier are concatenated is stored for each Chinese numeral that can be located immediately before the classifier. In this accent binding table,
Using the boundary between the numeral and the classifier as a reference position (0), information indicating which mora of the classifier has an accent nucleus when an accent combination occurs is represented by a number. Note that a negative value means that the accent nucleus moves leftward from the reference position.

[0006] However, the value PL of the accent position means that it is flat (no accent nucleus).
In addition, the reading is described for simplicity, and is not included in the information of the accent combination table.

Using this accent binding table,
It is possible to determine the position of the accent when a number and a classifier are concatenated. For example, 4 yen… Yo エ ン en (accent position “0”) 15 yen… Ju (↑) go ↓ en (accent position “0”) 100 yen… Hya (↑) cuen (accent position “PL”) 1,000 yen ... Sen-en (accent position "-1").

Similarly, when the numeral is "200" or "4000", the connected numerals are "100" and "1000", respectively. … Yo (↑) However, these accents are different from the original accent position, ni (↑) hyak ↓ enyo (↑) nsenen, and there was a problem that accurate accent positions could not be generated in some cases.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described problems. In determining an accent position at the time of connecting words, information of a word preceding a word at a combination position is used as necessary. It is an object of the present invention to provide a speech synthesis apparatus and method capable of generating a more natural accent.

Another object of the present invention is to make it possible to use information of a kanji numeral located before a kanji numeral immediately before the connection position in the connection between a numeral and a classifier. The object of the present invention is to make it possible to generate a more natural accent when connected.

[0011]

Means for Solving the Problems A speech synthesizing apparatus according to an aspect of the present invention for achieving the above object has, for example, the following configuration. That is, using a notation belonging to the first attribute and composed of at least one word and a notation belonging to the second attribute and composed of at least one word as keys, In speech-synthesizing a notation obtained by connecting a first notation part belonging to a first attribute and a second notation part belonging to a second attribute, for each of the first notation part and the second notation part, Extracting means for extracting the longest matching notation from the notations registered in the accent connection table as first and second keys, and using the first and second keys extracted by the extraction means to extract the accent connection table Means for retrieving an accent position at the time of connection, and a speech synthesis of the notation based on the notation to be speech-synthesized and the accent position acquired by the acquiring means. And a combining means for performing.

A method for controlling a speech synthesizer according to another embodiment of the present invention for achieving the above object has, for example, the following steps. That is, it belongs to the first attribute, is composed of at least one word, and belongs to the second attribute.
What is claimed is: 1. A method for controlling a speech synthesizer comprising an accent combination table for holding an accent position at the time of connection by using a notation composed of at least one word as a key, comprising: In the speech synthesis of the notation formed by connecting the second notation parts belonging to the attribute, the notation that is the longest match from the notation registered in the accent combination table for each of the first notation part and the second notation part Extracting the first and second keys as the first and second keys, and obtaining the accent connection table using the first and second keys extracted in the extracting step to obtain an accent position at the time of connection. And a synthesizing step of synthesizing the notation based on the notation to be synthesized and the accent position acquired in the acquiring step.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram showing the configuration of the speech synthesizer according to the present embodiment. In FIG. 1, reference numeral 21 denotes a control memory which stores a control program according to a control procedure as shown in the flowchart of FIG. Reference numeral 22 denotes a central processing unit that performs determination, calculation, and the like in accordance with a control procedure stored in the control memory 21. 23 is a memory (RA
M), and provides a work area when the central processing unit 22 performs various controls. The memory 23 includes a “numerical notation data holding unit 101”, a “standard Kanji numeral notation data holding unit 103”, and a “kanji numeral search key holding unit 10” described in FIG.
5, "a classifier search key holding unit 107", and a "classifier notation data holding unit 109".

Reference numeral 24 denotes a disk device, which uses a hard disk in this embodiment. The disk device 24 is shown in FIG.
Are stored in the memory 23 at the time of use. 25
Is a bus connecting the above components.

FIG. 2 is a block diagram showing a functional configuration of the speech synthesizer according to the present embodiment. In FIG.
Numeral notation data holding unit for holding the notation of a numeral, 102
Is a standard kanji notation conversion processing unit for converting the notation of the numeral into the standard kanji notation, 103 is a standard kanji notation data holding unit for holding the result of conversion by the standard kanji conversion processing unit, and 104 is a standard kanji notation data holding unit. A kanji numeral search key generation unit 105 for generating a search key for an accent combination table from numeral notation data, and a kanji numeral search key holding unit 105 for holding the processing result of the kanji numeral search key generation unit.

Reference numeral 106 denotes an accent connection table search processing unit for searching the accent connection table 110 by using a Chinese numeral search key and a classifier search key. 107 is a classifier search key holding unit for holding a classifier search key generated from the classifier notation data, 108 is a classifier search key generation processing unit for generating a classifier search key from the classifier notation data, and 109 is a classifier holding the classifier notation data The notation data holding unit 110 is an accent binding table that holds accent position values corresponding to the Chinese numeric search key and the classifier search key.

Next, the procedure of the speech synthesis processing by each of the above-described functional configurations will be described with reference to the flowchart shown in FIG. First, in step S301, the standard kanji notation conversion processing unit 102 uses the notation of the numeral stored in the numeral data storage unit 101 (eg, 123, 14,400, 1,500).
Is converted to standard Chinese numeral notation (e.g., 123, 14,400, 1,500) to eliminate the fluctuation of the notation, and the conversion result is stored in the standard Chinese numeral notation data storage unit 103. Then, control goes to a step S302.

In step S302, the kanji numeric search key generation processing unit 104 sets the standard kanji numeric notation data holding unit 103
Generates a kanji search key required for searching the accent binding table from the standard kanji notation stored in. The generated Chinese numeral search key is stored in the Chinese numeral search key storage unit 105.

Since the Chinese numeral search key depends on the data structure of the accent combination table 110, the accent combination table 110 will be described here. The accent combination table 110 has a matrix structure in which Chinese numbers are written in rows and classifiers are written in columns, and values of accent positions are held at intersections of rows and columns. FIG. 4 shows a data configuration example of the accent combination table according to the present embodiment. The numbers representing the accent positions in FIG. 4 are defined such that the boundary between the numeral and the classifier is 0, the right direction (the direction of the classifier) is plus, the left direction (the direction of the numeral) is negative, and the position of the accent nucleus from the connection position It is a numerical value indicating whether it is in Mora. PL is a symbol indicating that the plate is of a flat plate type (without accent nuclei).

The kanji numeric search key generation processing unit 104 starts with the first digit of the standard kanji notation, and
The longest partial character string that matches the Chinese numeral notation of 0 is used as a Chinese numeric search key, and is stored in the Chinese numeric search key storage unit 105. After the generation of the Chinese numeral search key, the process proceeds to step S303.

Next, in step S303, the classifier search key generation processing unit 108 sets the classifier notation data holding unit 109.
A classifier search key is generated from the classifier notation held in. Accent binding table 11 as shown in FIG.
In the case of 0, the notation of the classifier becomes the classifier search key as it is. When the classifier search key is generated as described above, the classifier search key is used as the classifier search key holding unit 107.
To be held. Thus, the process proceeds to step S304.

In step S304, the accent binding table search processing unit 106 sets the
Based on the Chinese numeric search key held in No. 5 and the classifier search key held in the classifier search key holding unit 107, the accent binding table 110 is searched, and the corresponding value (value indicating the accent position) is searched. Is output, and all processing ends.

As a result of the above processing, for example, for "200 yen", the Chinese numeral search key is "200", the classifier search key is "yen", and the accent position table 110 in FIG. "0" is output. As a result, at the time of speech synthesis, an accent position of "Ni (↑) Hyaku ↓ En" is obtained. That is, a correct accent position can be obtained.

Similarly, for "4,000 yen", the Chinese numeral search key is "4,000", the classifier search key is "yen", and the accent position "PL" is output from the accent combination table 110 of FIG. . As a result, during speech synthesis,
The correct accent position "Yo (↑) Neng Seng En" is obtained.

In the above embodiment, the case where the notation of a classifier is used as a classifier search key of the accent combination table 110 has been described, but the present invention is not limited to this. For example, considering the reduction in table size,
The classifiers having the same value of the moving mora number of the accent nucleus with respect to the Chinese numeral search key may be collected in one category. In this case, a dictionary or a correspondence table holding category information of classifiers is prepared.
Then, the classifier search key generation processing unit 108 extracts classifier category information from the category information dictionary or the correspondence table, and generates a classifier search key.

Further, in the above embodiment, the case where the accent combination table 110 is realized on the memory 23 (RAM) has been described. However, the present invention is not limited to this, and is realized by using an arbitrary storage medium. Is also good.

Further, in the above embodiment, the case where each unit is configured on the same computer has been described. However, the present invention is not limited to this, and each unit is divided into computers and processing devices distributed on a network. You may comprise.

As described above, according to the present embodiment, the position of the accent when a number and a classifier are concatenated can be obtained as follows. That is, after converting a numeral into a standard Chinese numeral notation, a partial character string necessary for searching the accent combination table is sequentially cut out from the first place of the standard Chinese numeral notation, and is used as a Chinese numeral search key. Also,
A classifier or a classifier is used as a classifier search key.
Then, the accent combination table is searched using these two search keys, and the position of the accent when the number and the classifier are connected is determined. For this reason, it is possible to determine an accent position in consideration of a numeral word located before a numeral word immediately before a connection position, and to generate a more accurate accent,
As a result, a natural synthesized speech can be generated.

The present invention can be applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), but it can be applied to a single device (for example, a copier, a facsimile). Device).

Another object of the present invention is to provide a storage medium storing a program code of software for realizing the functions of the above-described embodiments to a system or apparatus, and to provide a computer (or CPU) of the system or apparatus.
And MPU) read and execute the program code stored in the storage medium.

In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention.

As a storage medium for supplying the program code, for example, a floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, CD
-R, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

When the computer executes the readout program code, not only the functions of the above-described embodiment are realized, but also the OS (Operating System) running on the computer based on the instruction of the program code. ) May perform some or all of the actual processing, and the processing may realize the functions of the above-described embodiments.

Further, after the program code read from the storage medium is written into a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, based on the instructions of the program code, It goes without saying that the CPU included in the function expansion board or the function expansion unit performs part or all of the actual processing, and the processing realizes the functions of the above-described embodiments.

[0036]

As described above, according to the present invention, when determining the accent position at the time of connecting words, the information of the word preceding the word at the connecting position is used as necessary. Natural accents can be generated.

According to the present invention, in particular, in the connection between a numeral and a classifier, it is also possible to use information on a Chinese numeral located before a Chinese numeral immediately before the connection position. When connected, a more natural accent can be generated.

[0038]

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech synthesis device according to an embodiment.

FIG. 2 is a block diagram illustrating a functional configuration of the speech synthesizer according to the embodiment;

FIG. 3 is a flowchart illustrating a procedure of an accent position determination process in the speech synthesizer of the embodiment.

FIG. 4 is a diagram illustrating an example of a data configuration of an accent combination table according to the embodiment;

FIG. 5 is a diagram showing a general accent combination table of a classifier "en" and a numeral.

Claims (12)

[Claims] 1. An accent that holds a position of an accent at the time of connection using a notation belonging to at least one word belonging to a first attribute and a notation belonging to at least one word belonging to a second attribute as keys. In speech-synthesizing a combined table and a notation formed by connecting a first notation part belonging to a first attribute and a second notation part belonging to a second attribute, the first notation part and the second notation part are combined. Extracting means for extracting, as first and second keys, the longest matching notation from the notations registered in the accent binding table for each of the notations, using the first and second keys extracted by the extracting means; Acquiring means for retrieving an accent connection table and acquiring an accent position at the time of connection; and a sound of the notation based on the notation to be synthesized and the accent position acquired by the acquiring means. A voice synthesizing device comprising voice synthesis means for performing voice synthesis. 2. The method according to claim 2, wherein the first attribute is a numeral, and the second attribute is
2. The speech synthesizer according to claim 1, wherein the attribute is a classifier. 3. The method according to claim 1, wherein the extracting means extracts the longest matching notation from the lower digit side of the numeral in the first notation portion from the accent binding table as a first key, and
The speech synthesizer according to claim 2, wherein a notation that matches a classifier in a notation portion is extracted from the accent connection table and used as a second key. 4. The speech synthesizing apparatus according to claim 2, further comprising a conversion unit that converts the notation of the first notation unit into a standard Kanji number notation and provides the converted notation to the extracting unit. 5. The accent binding table classifies classifiers having the same accent position in the combination of all registered numeral expressions and classifiers into the same category, and uses the category or classifier as a key, and The method according to claim 1, further comprising: storing an accent position based on a concatenation of a numeral and a classifier, wherein the extracting unit extracts, as a second key, a notation or a category corresponding to the classifier in the second notation unit from the accent combination table. 3. The speech synthesizer according to 3. 6. An accent that holds a position of an accent at the time of connection using a notation belonging to at least one word belonging to the first attribute and a notation belonging to at least one word belonging to the second attribute as keys. A method for controlling a speech synthesizer provided with a connection table, wherein speech synthesis is performed on a notation formed by connecting a first notation part belonging to a first attribute and a second notation part belonging to a second attribute. For each of the first notation part and the second notation part, an extraction step of extracting the longest matching notation from the notations registered in the accent binding table as first and second keys, and extracted in the extraction step. Using the first and second keys to search the accent binding table to obtain an accent position at the time of connection; A synthesizing step of synthesizing the notation based on the xent position. 7. The method according to claim 1, wherein the first attribute is a numeral, and the second attribute is
7. The method according to claim 6, wherein the attribute is a classifier. 8. The method according to claim 1, wherein the extracting step extracts the longest matching notation from the lower digit side of the numeral in the first notation part from the accent binding table as a first key.
8. The method according to claim 7, wherein a notation that matches a classifier in a notation unit is extracted from the accent binding table and used as a second key. 9. The control of the speech synthesizer according to claim 7, further comprising a conversion step of converting the notation of the first notation unit into a standard Chinese numeral notation and providing the converted notation to the extraction step. Method. 10. The accent binding table classifies classifiers having the same accent position in the combination of all registered notation of a number and a classifier into the same category, and uses the category or the classifier as a key, and The method according to claim 1, further comprising: storing an accent position based on a concatenation of a number and a classifier, wherein the extracting step extracts, as a second key, a notation or a category matching the classifier in the second notation unit from the accent combination table. 9. The method for controlling a speech synthesizer according to claim 8. 11. An accent that holds a position of an accent at the time of concatenation using a notation belonging to at least one word belonging to a first attribute and a notation belonging to at least one word belonging to a second attribute as keys. A computer readable memory storing a control program for a speech synthesizer using a connection table, wherein the control program is configured by connecting a first notation part belonging to a first attribute and a second notation part belonging to a second attribute. In the speech synthesis of the notation consisting of: extracting the longest matching notation from the notations registered in the accent combination table as the first and second keys for each of the first notation portion and the second notation portion The accent binding table is searched using the code of the process and the first and second keys extracted in the extraction process, and the accent position at the time of connection is obtained. A computer-readable memory, characterized in that it comprises a code acquisition step, and a code combining step of performing speech synthesis of the title based on the accent position denoted said to be speech synthesized acquired in the acquisition step of. 12. The computer readable memory according to claim 11, further comprising storing the accent binding table.