JP2001014310A - Device and method for compressing conversion dictionary used for voice synthesis application - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate a new dictionary wherein the priority of data is set according to the needs of users. SOLUTION: This method includes a dictionary size input part for inputting the size of a dictionary and a priority decision part 12 which decides the priority of vocabularies stored in the dictionary. The priority decision part extracts vocabularies of high priority from a dictionary main body 13 in order according to corpus 14 information, and stores extracted vocabularies up to the dictionary size inputted at the dictionary input part 11 to generate a new dictionary.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and a method for compressing a dictionary used for conversion in an application for performing speech synthesis based on text data, waveform data, and the like.

[0002]

2. Description of the Related Art Due to the rapid progress of computer technology in recent years, there have been many applications using voice, which have been shunned because the processing time is not within a practical range due to the huge amount of data. It is being created.

[0003] Above all, applications that artificially synthesize speech based on text data, waveform data, and the like are widely used to realize a user interface that does not make the user aware of using a computer. It is being used.

In order to realize such an application, a dictionary for conversion is essential. That is,
In a speech synthesis application, a word dictionary and a waveform dictionary are required.

However, in general, in order to increase the clarity of speech synthesis using these dictionaries, the information to be included in the dictionaries may be enormous, and the more the amount of information, the longer the processing time becomes. Because it becomes long,
Determining how much information should be included in a dictionary is an important practical issue.

Many methods have been disclosed for creating a practical and compact dictionary. For example,
Japanese Patent Application Laid-Open No. 5-189415 discloses a method of creating a new selected dictionary by extracting basic words based on long-term learning information and the activation value of a neuro table from a dictionary having a fixed priority. I have.

[0007]

However, in the conventional method as described above, since the priorities of the original dictionaries are fixed, depending on the basic word extraction conditions for creating a new selected dictionary. However, there is a problem that the extraction of the basic word may not be performed efficiently. In other words, the needs of the user change from moment to moment, and the fact that the priorities are fixed means that the search order in the dictionary is fixed even if extraction conditions are added by long-term learning information or the like. Since the order is followed, the search efficiency of the basic word to be extracted is low depending on the added extraction condition, and the extraction may take time.

[0008] In addition, when extraction conditions due to special circumstances are required, for example, vocabulary relating to a specific magazine is preferentially extracted, or a priority criterion such as this category for today, this category for tomorrow, and the like. In the case of frequently changing in a short period, the conventional method as described above cannot cope with the situation.

In order to solve the above problems, the present invention provides a conversion dictionary compression apparatus and method for use in a speech synthesis application capable of creating a new dictionary in which data priorities are set according to the needs of the user. The purpose is to provide.

[0010]

According to the present invention, there is provided a conversion dictionary compression apparatus used for a speech synthesis application according to the present invention, comprising: a dictionary size input unit for inputting a dictionary size; and a vocabulary stored in the dictionary. A priority determining unit that determines the priority of the dictionary, and in the priority determining unit, extracts words having a high priority from the dictionary body in order based on the corpus information, and obtains the dictionary size input by the dictionary size input unit. A new dictionary is created by storing the extracted vocabulary up to this point.

With this configuration, the priority of the vocabulary included in the dictionary can be dynamically changed in accordance with the corpus information. Therefore, the vocabulary related to a specific magazine is preferentially extracted, or the priority standard is frequently set in a short period of time. , It is possible to efficiently search for basic words to be extracted simply by changing the corpus information.

Next, in order to achieve the above object, a conversion dictionary compression method used for a speech synthesis application according to the present invention comprises the steps of inputting the size of a dictionary and determining the priority of vocabulary stored in the dictionary. And extracting a vocabulary having a high priority from the dictionary body based on the corpus information, and storing the extracted vocabulary until the input dictionary size is reached, to create a new dictionary.

With this configuration, the priority of the vocabulary included in the dictionary can be dynamically changed according to the corpus information. Therefore, the vocabulary related to a specific magazine can be preferentially extracted, or the priority criterion is frequently set in a short period of time. , It is possible to efficiently search for basic words to be extracted simply by changing the corpus information.

Further, the present invention is characterized by software for executing the function of the conversion dictionary compression used for the above-mentioned speech synthesis application as a processing step of a computer, and more specifically, to input a dictionary size. And deciding the priority order of the vocabulary stored in the dictionary. The vocabulary with the highest priority is sequentially extracted from the dictionary body based on the corpus information, and extracted until the input dictionary size is reached. A conversion dictionary compression method used for a speech synthesis application that stores a vocabulary and creates a new dictionary, and a computer-readable recording medium that records such a process as a program.

With this configuration, by loading and executing the program on the computer, the priority of the vocabulary included in the dictionary can be dynamically changed in accordance with the corpus information. Conversion dictionary compression for speech synthesis applications that enables efficient search of basic words to be extracted by simply changing corpus information, even when the priority criteria frequently change in a short period of time or when the priority criteria change frequently. The device can be realized.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a conversion dictionary compression device used for a speech synthesis application according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a conversion dictionary compression device used for a speech synthesis application according to an embodiment of the present invention. In FIG. 1, 1
1 is a dictionary size input unit, 12 is a priority order determination unit, and 1 is a dictionary size input unit.
3 indicates a dictionary body, 14 indicates a corpus, and 15 indicates a newly created dictionary.

In FIG. 1, first, a dictionary size input unit 11
Enter the size of the dictionary created in. The dictionary size is entered, for example, by using a PDA (Personal Digital
This is because it is possible to use an application using voice even in a medium having limited computer resources such as a sistant) and a portable terminal. Therefore,
Here, the dictionary size to be input specifies a value smaller than the original dictionary size. The input need not be limited to only the size of the dictionary, but may be the number of words registered in the dictionary.

Then, the priority order judging unit 12 extracts the basic words in descending order of priority until the size of the dictionary input from the dictionary size input unit 11 is reached while referring to the dictionary main body 13. Here, the priority itself is corpus 1
4 is determined based on the content. The contents of the corpus 14 may be, for example, documents related to “computers” or contents published in a specific newspaper.

Finally, while the basic words that match the basic word extraction conditions are sequentially extracted from the dictionary body 13, the newly created dictionary 1
5 is created. The newly created dictionary 15 created in this way is smaller in dictionary size than the dictionary itself, whereas the clarity of synthesized speech in a speech synthesis application, for example, under specific conditions, It is possible to compare with using the main body. For example, when a newly created dictionary 15 in which the basic words belonging to the category “computer” are preferentially extracted by the corpus 14 is used, the ordinary dictionary 13 is used for the synthesized speech of the words related to “computer”. It can be expected that the same clarity is maintained as in the case where the image is displayed.

FIG. 2 shows an embodiment in which the conversion dictionary compression device used for the speech synthesis application according to the embodiment of the present invention is applied to a speech synthesis application.
In FIG. 2, 21 is an input device, 22 is a priority determination device, 23 is a dictionary body, 24 is a corpus, 25 is a newly created dictionary, 26 is a waveform dictionary, and 27 is a speech synthesis application. Show.

In FIG. 2, first, a compression size of a dictionary to be used in the input device 21 is inputted. The input need not be limited to only the size of the dictionary, and the number of registered words in the dictionary may be input.

Then, in the priority determining device 22,
While referring to the dictionary main body 23, the basic words are extracted in descending order of priority so as to have the dictionary size input from the input device 21. Here, the priority itself can be freely set by the user by using the corpus 24. For example, the priority of basic words belonging to the category of "Internet" may be increased, or the priority of basic words published in a specific magazine or trade paper may be increased.

Finally, while the basic words that match the basic word extraction conditions are sequentially extracted from the dictionary body 23, the newly created dictionary 2
5 is created. A speech synthesis application 27 is formed based on the newly created dictionary 25 and the waveform dictionary 26.

The newly created dictionary 2 created in this way
5 indicates that the dictionary size is smaller than that of the dictionary itself, whereas the clarity of the synthesized speech in a speech synthesis application, for example, is not inferior to using the dictionary body under specific conditions. It is possible. For example, when a newly created dictionary 25 in which the basic words belonging to the category of “Internet” are preferentially extracted by the corpus 24, the synthesized dictionary of the words related to “Internet” is stored in the normal dictionary 2.
It can be expected that the same clarity as in the case of using No. 3 is maintained.

Further, the configuration may be such that the priority itself can be directly changed and input. For example, FIG. 3 shows an embodiment in which a priority changing device 31 is added to the conversion dictionary compression device used for the speech synthesis application according to the embodiment of the present invention.

In FIG. 3, the priority changing device 31
Parameters such as the number of appearances can be changed, while data itself to be stored in the dictionary can be changed. That is, the priority distribution of the dictionary itself can be changed by changing parameters such as the number of appearances, and the number of appearances can be changed by changing (adding, deleting, etc.) the data itself to be stored in the dictionary. Changes, and the priority of the dictionary is changed.

FIG. 4 shows an embodiment in which the same dictionary size compression processing is performed on the waveform dictionary. Here, the waveform dictionary means an output waveform of the voice itself. Therefore, even if the same word "a" is displayed, the output waveform changes depending on the characters before and after the word or the context.

In FIG. 4, first, the compression size of the waveform dictionary 26 to be used in the input device 21 is input. It is not necessary to limit only to the size of the waveform dictionary.
The upper limit of the number of registered waveforms in the waveform dictionary may be input, or a category of a waveform dictionary to be newly created may be input.

Then, in the priority determining device 22,
While referring to the waveform dictionary main body 26, waveforms are extracted in descending order of priority so as to have the dictionary size input from the input device 21. Here, the priority itself is the corpus 24
Can be set freely by the user. For example, the priority of a waveform belonging to the category "Internet" may be increased, or the priority of a voice waveform expressed in a specific interview or the like may be increased.

Finally, while sequentially extracting the waveforms meeting the waveform extraction conditions from the waveform dictionary main body 26, the new waveform dictionary 4
Create one. The new waveform dictionary 41 and the dictionary 23
, A speech synthesis application 27 is formed.

The new waveform dictionary 4 thus created
1 indicates that the size of the dictionary is smaller than that of the waveform dictionary main body 26, whereas the clarity of synthesized speech in a speech synthesis application, for example, is that the waveform dictionary main body 26 is used under specific conditions. It is possible to make it comparable.

As shown in FIG. 5, when a new dictionary is created, it can be created while checking the coverage. Here, the cover rate is an index indicating what percentage of the words used in the sentence are covered by the words registered in the newly created dictionary when the sentence is created.
In FIG. 5, reference numeral 51 denotes a cover ratio calculating device, and 52 denotes a cover ratio display device.

When the newly created dictionary 25 is created, the coverage calculator 51 refers to the dictionary with the original dictionary body 23, and
An objective index indicating how much the newly created dictionary 25 can cover is calculated. Then, the calculated value is always displayed on the coverage ratio display device 52.

As described above, the conditions input from the input device 21 and the contents of the corpus 24 can be changed while monitoring the coverage in the newly created dictionary 25. It is possible to reduce the dictionary size as much as possible.

FIG. 6 shows an embodiment for creating a new waveform dictionary while confirming the clarity when the size compression target is not a dictionary but a waveform dictionary. In other words, when a new waveform dictionary is created by reducing the waveform data, the clarity of the finally output synthesized speech is clarified, so that the size of the waveform dictionary is reduced to the limit that can be used by the user. Is what you do. In FIG. 6, 6
Reference numeral 1 denotes a clarity estimation device, and 62 denotes a clarity display device.

In FIG. 6, when a new waveform dictionary 41 is created, the clarity estimating device 61 is created together with the original waveform dictionary main body 26.
Calculates an objective index indicating to what extent the synthesized speech is clear by the new waveform dictionary 41. Then, the calculated value is always displayed on the clarity display device 62.

As described above, the conditions input from the input device 21 and the contents of the corpus 24 can be changed while monitoring the clarity of the new waveform dictionary 41. The size of the waveform dictionary can be reduced as much as possible.

As described above, according to the present embodiment, in a speech synthesis application, the dictionary size for conversion can be compressed without lowering the conversion efficiency.

Next, a description will be given of a processing flow of a program for realizing the conversion dictionary compression apparatus used for the speech synthesis application according to the embodiment of the present invention. FIG. 8 is a flowchart showing the processing of a program for realizing the conversion dictionary compression device used for the speech synthesis application according to the embodiment of the present invention.

In FIG. 7, the user inputs a dictionary size as one of the conditions for compressing the dictionary (step S70).
1). Then, the basic words are extracted from the dictionary body in descending order of priority (step S702).

The basic words extracted from the dictionary body are sequentially stored in the newly created dictionary as data constituting a new dictionary (step S703). When the newly created dictionary has the dictionary size requested by the user in our office (step S704: Yes), the newly created dictionary can be created as a dictionary for an application.

As shown in the example of the recording medium shown in FIG. 8, the recording medium storing the program for realizing the conversion dictionary compression apparatus used for the speech synthesizing application according to the embodiment of the present invention is a CD-ROM 82-1. Not only a portable recording medium 82 such as a floppy disk 82-2, but also any other storage device 81 provided at the end of a communication line or a recording medium 84 such as a hard disk or a RAM of a computer 83 may be used. , The program is loaded and executed on the main memory.

Also, as shown in the example of the recording medium shown in FIG. 8, a recording medium that records a newly created dictionary and the like generated by the conversion dictionary compression device used for the speech synthesis application according to the embodiment of the present invention is CD-ROM 82-1
Recording medium 82 such as a disk or a floppy disk 82-2
Not only other storage devices 8 provided at the end of the communication line
1 or a recording medium 84 such as a hard disk or a RAM of the computer 83, which is read by the computer 83 when using the conversion dictionary compression apparatus used for the speech synthesis application according to the present invention.

[0044]

As described above, according to the conversion dictionary compression apparatus used in the speech synthesis application according to the present invention, the priority of words included in the dictionary can be dynamically changed according to the corpus information. Even if the vocabulary relating to the magazine is preferentially extracted or the priority standard changes frequently in a short period of time, it is possible to efficiently search for the basic words to be extracted.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a conversion dictionary compression device used for a speech synthesis application according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of a conversion dictionary compression device used for a speech synthesis application according to an embodiment of the present invention;

FIG. 3 is a configuration diagram of a conversion dictionary compression device used for a speech synthesis application according to another embodiment of the present invention.

FIG. 4 is a configuration diagram of a conversion dictionary compression device used for a speech synthesis application according to another embodiment of the present invention.

FIG. 5 is a configuration diagram of a conversion dictionary compression apparatus used for a speech synthesis application according to another embodiment of the present invention.

FIG. 6 is a configuration diagram of a conversion dictionary compression device used for a speech synthesis application according to another embodiment of the present invention.

FIG. 7 is a flowchart of processing in the conversion dictionary compression device used for the speech synthesis application according to the embodiment of the present invention;

FIG. 8 is an exemplary diagram of a recording medium.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 11 Dictionary size input part 12 Priority judgment part 13, 23 Dictionary main body 14, 24 Corpus 15, 25 Newly created dictionary 21 Input device 22 Priority judgment device 26 Waveform dictionary 27 Voice synthesis application 31 Priority change device 41 New waveform dictionary 51 Coverage ratio calculation device 52 Coverage ratio display device 61 New waveform dictionary 61 Clarity estimation device 62 Clarity display device 81 Storage device of line destination 82 Portable recording medium such as CD-ROM or floppy disk 82-1 CD-ROM 82- 2 Floppy disk 83 Computer 84 Recording media such as RAM / hard disk on computer

Claims (3)

[Claims] 1. A dictionary size input unit for inputting the size of a dictionary, and a priority determination unit for determining a priority of a vocabulary stored in the dictionary, wherein the priority determination unit determines a priority of a vocabulary based on corpus information. A speech synthesis application characterized by sequentially extracting vocabularies of high priority from the dictionary body, storing the extracted vocabulary until the dictionary size input by the dictionary size input unit and creating a new dictionary. Conversion dictionary compression device to be used. 2. A step of inputting a size of the dictionary, and a step of determining a priority of a vocabulary stored in the dictionary, extracting words having a higher priority from the dictionary body in order based on the corpus information, A conversion dictionary compression method for use in a speech synthesis application, wherein a new dictionary is created by storing the vocabulary extracted until the input dictionary size is reached. 3. A step of inputting a size of the dictionary, and a step of determining a priority of a vocabulary stored in the dictionary, extracting words having a higher priority from the dictionary body in order based on the corpus information, A computer-readable recording medium storing a program to be executed by a computer, storing the vocabulary extracted until the input dictionary size is reached, and creating a new dictionary.