JPH05108086A - Rhythm symbol generation device - Google Patents

Abstract

PURPOSE:To speedily and accurately generate a rhythm symbol through easy operation even when a large-quantity document is processed. CONSTITUTION:When a document read part 1 reads in a document to be analyzed, a morpheme dictionary looking-up part 3 refers to a morpheme dictionary 2 for character strings of the read-in document, thereby obtaining node data as a morpheme analysis result. A connection certification part 5 certifies a connection between node data by referring to a connection table 4 and generate a morpheme lattice wherein the connection relation between morphemes and its score (extent of likelihood of connection) are entered. A solution generation part 6 stores the generated morpheme lattice in an analysis result storage part 7 up to specific order as an analysis result. An analysis result selection part 8 selects the most desirable analysis result among analysis results in the analysis result storage part 7 and a rhythm generation part 9 generates rhythm according to the selected analysis result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a prosody symbol generation device for generating a prosody symbol used for speech synthesis or the like.

[0002]

2. Description of the Related Art Generally, in the case of performing speech synthesis, a prosody symbol generation device is used which analyzes a character string and regularly generates prosody symbols based on the analysis result. In this prosody symbol generation device, it is necessary to perform a correct morphological analysis, a correct dependency analysis, etc. in order to obtain a correct prosody.

[0003]

However, in the conventional prosodic symbol generation device, the correct morphological analysis and the correct dependency analysis are performed by the operation of the operator. When it is necessary to generate prosodic symbols, the operation becomes extremely complicated and there is a drawback that a large amount of documents cannot be processed by a simple operation.

The present invention, even when processing a large number of documents,
It is an object of the present invention to provide a prosody symbol generation device capable of quickly and accurately generating a prosody symbol with a simple operation.

[0005]

In order to achieve the above object, the present invention provides a morphological analysis means for performing morphological analysis of a character string, and an analysis result storage in which the analysis result of the character string is stored with a likelihood. Means and an analysis result selection means for selecting the analysis result stored in the analysis result storage means by referring to the likelihood, and generating a prosody symbol based on the selected analysis result. It is characterized by

[0006]

In the prosody symbol generation apparatus of the present invention, the analysis result obtained by performing the morpheme analysis of the character string is stored with the likelihood in the analysis result storage means. For example, when the user selects the analysis result stored in the analysis result storage means by the analysis result selection means with reference to the likelihood, a prosody symbol is generated based on the selected analysis result. As described above, according to the present invention, the user can select the analysis result by the analysis result selecting means, so that even a large number of documents can be processed by a simple operation.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an embodiment of a prosody symbol generation apparatus according to the present invention. Referring to FIG. 1, the prosody symbol generation apparatus according to the present exemplary embodiment refers to a document reading unit 1 that reads a document to be analyzed, and a morpheme dictionary 2 for a character string of the read document. The morpheme dictionary drawing unit 3 for obtaining data and the connection between the node data are tested with reference to the connection table 4,
A connection test unit 5 that generates a morpheme lattice that describes the connection relations between morphemes and their scores (that is, the degree of likelihood of being connected: likelihood); Is generated as an analysis result, an analysis result storage unit 7 that stores the analysis results (generally a plurality) extracted by the solution generation unit 6, and an analysis result stored in the analysis result storage unit 7. Of these, an analysis result selection unit 8 that selects the most desirable analysis result, and a prosody generation unit 9 that generates a prosody based on the selected analysis result
And have.

FIG. 2 is a diagram showing an example of the configuration of the analysis result selection unit 8. The analysis result selection unit 8 basically reads the analysis result from the analysis result storage unit 7 in accordance with the user's operation. It is configured to be presented to the user and to be selected by the user. To this end, FIG.
The analysis result selection unit 8 of is a selection instruction unit 11 for allowing the user to perform various selection instructions, and allows the user to specify a predetermined word in the presented analysis result, and based on the specified word. A word designation unit 12 that narrows down the analysis result in the analysis result storage unit 7, and a user designates a predetermined word division position in the presented analysis result, and the analysis result storage unit 7 is based on the designated word division position. And a word delimiter position designating unit 13 for narrowing down the analysis result in.

The selection instructing unit 11 determines the automatic reading switch SW1 for instructing the user whether or not to operate the analysis result selecting unit 8 and / or the currently selected analysis result as the correct answer. Then, a next sentence switch SW2 for reading the next character string is provided. The automatic reading switch SW1 and the next sentence switch S
When both W2 are provided, the selection instructing unit 11
Does not wait for the operator's designation when the automatic reading switch SW1 is specified, processes the next character string with the first solution in the analysis result storage section as the correct answer, and specifies the automatic reading switch. If not, the next character string is not processed until the next sentence switch SW2 is designated.

Further, the selection instruction section 11 presents the next candidate switch SW3 for presenting the most likely analysis result next to the currently selected analysis result, and / or presents it before the currently selected analysis result. A front candidate switch SW4 for presenting the analyzed result again is provided.

Further, when the user designates an incorrect word in the presented analysis result, the word designating section 12
The analysis result including the erroneous word is deleted from the analysis result storage unit 7, and the analysis result to be presented next is selected from the remaining analysis results, or the analysis result presented by the user is selected. When a correct word is designated, the analysis result that does not include the correct word is deleted from the analysis result storage unit 7, and the analysis result to be presented next is selected from the remaining analysis results.

Further, when the user specifies a correct word break position in the presented analysis results, the word break position specifying unit 13 stores the analysis result in which the word break does not exist at the specified position. Deleted from the section 7, select the analysis result to be presented next from the remaining analysis results, or when the user specifies the position of the incorrect word break in the presented analysis result, The analysis result having the word break at the designated position is deleted from the analysis result storage unit 7, and the analysis result to be presented next is selected from the remaining analysis results.

Next, the operation of the prosody symbol generating apparatus having such a configuration will be described. The document reading unit 1 reads, for example, a Japanese document, examines a Japanese character string in this document, and if there is a Japanese character string, divides it into units for morphological analysis and uses the morphological dictionary lookup unit 3 to analyze characters. Pass the row. When a character string is input to the morpheme dictionary lookup unit 3, the morpheme dictionary lookup unit 3 refers to the morpheme dictionary 2 for each part of the character string, obtains node data having morpheme grammar and position information, and Is given to the connection verification unit 5.

The connection verification unit 5 verifies the validity of the connection between the morphemes with reference to the connection table 4 based on the node data, that is, the analysis result of the morphemes. The connection table 4 describes the score of the connection when one part of speech is followed by another part of speech, that is, the likelihood, and the score is "0" when the connection is not valid. When the score is not “0” and is a predetermined value with the validity of the connection, the connection verification unit 5 intends to connect the preceding and following node data, and connects the respective node data. Write down your opponent and connection scores.

As a result of the processing up to this point, a morpheme lattice in which the connection relation between morphemes and its score (likelihood) are described is generated. FIG. 3 is a diagram showing an example of the morpheme lattice generated in this way, in which [] indicates the connection score.

After that, the solution generation unit 6 traces the morpheme lattices generated as described above and averages the scores of the paths up to the end of the sentence by the number of nodes. Is sent to the analysis result storage unit 7 as an analysis result and stored therein. First, the analysis result selection unit 8 presents the operator with a solution having the highest average score among the analysis results stored in the analysis result storage unit 7. For example, this presentation is performed through the prosody generation unit 8,
This is done by causing the voice synthesizer 10 to output a voice.

At this time, when the automatic reading switch SW1 is selected, an "automatic reading" signal is added to the analysis result selection section 8, whereby the analysis result selection section 8 has the highest average score. A candidate with a high value is automatically determined as a solution, and the document reading unit 1 is instructed to automatically read the next Japanese character string. In addition, this operation is
It is automatically performed until the automatic reading switch SW1 is turned off. On the other hand, when the automatic reading switch SW1 is not provided, or when the automatic reading switch SW1 is provided and is not selected, the operator can select a desired candidate.

That is, when the current solution is presented,
The operator can analyze whether the current solution is desirable. As a result, when a certain morpheme is a wrong candidate, the operator can specify the wrong word by the word specifying unit 12. When the wrong word is specified, the analysis result selection unit 8 deletes the solution including the specified morpheme from the analysis results in the analysis result storage unit 7, and has the highest score among the remaining solutions. Present candidates. On the contrary, when a certain morpheme is a correct candidate, the operator can also specify it by the word specifying unit 12. In this case, the analysis result selection unit 8 deletes the solution that does not include the designated morpheme from the analysis results in the analysis result storage unit 7, and selects the candidate with the highest score among the remaining solutions. Present.

Further, the operator can specify the position of the break of the word by the word division position specifying unit 13. When this position is designated, the analysis result selection unit 8
Of the analysis results in the analysis result storage unit 7, the solution having no word break at the specified position is deleted, and the candidate with the highest score among the remaining solutions is presented. On the contrary, the operator can specify the position which does not become the break of the word by the word division position specifying unit 13. When this position is designated, the analysis result selection unit 8 causes the analysis result storage unit 7 to
Among the analysis results in, the solution with a word break at the specified position is deleted, and the candidate with the highest score among the remaining solutions is presented.

At such a stage, the operator can operate the next candidate switch SW3 and the previous candidate switch SW4 to select a candidate. For example, when the next candidate switch SW3 is operated, the analysis result selection unit 8
Of the analysis results in the analysis result storage unit 7 presents the candidate having the next highest score to the candidate currently presented. On the other hand, when the previous candidate switch SW4 is operated, the analysis result selection unit 8 presents the candidate having the next lowest score among the analysis results in the analysis result storage unit 7 after the currently presented candidate. This allows easy selection of candidates.

In this way, when the operator wants to select a desired candidate and proceed with the analysis of the next Japanese character string,
The operator can operate the next sentence switch SW2.

That is, when the next-sentence switch SW2 is operated, a signal "read next sentence" is added to the analysis result selection unit 8, whereby the analysis result selection unit 8 corrects the currently presented candidate. Then, the prosody generation unit 9 generates a prosody symbol from the solution and stores it in the prosody file.
Then, the document reading unit 1 is instructed to read the next Japanese character string.

FIG. 4 is a diagram showing a concrete system configuration example of the prosody symbol generation apparatus of the present embodiment. In the example of FIG.
The prosody symbol generation device includes a control device 21 for controlling the whole,
It is realized by a storage device 22, a mouse 23, and a display 24. That is, the document data to be analyzed is stored in the storage device 22, and the analysis result storage unit 7 and the prosody file are also set in the storage device 22. In addition, the display 24 displays the solution selected from the analysis result storage unit 7.

When the prosody symbol generator of this embodiment is realized by the system shown in FIG. 4, the operator operates the mouse to move the cursor CS appearing on the display, and clicks the left and right mouse buttons. Sends an instruction to the analysis result selection unit 8.

That is, in the example of FIG. 4, when the shape of the cursor CS is “∘”, for example, when the left button of the mouse 23 is clicked, information that the morpheme at the position where the cursor CS exists is correct is displayed in the analysis result selection unit 8 Sent to. On the other hand, for example, if you click the right button, the cursor C
Information that the morpheme of the position where S exists is erroneous is sent to the analysis result selection unit 8.

In such a state, the display 2
Cursor CS to the place where "cut" is displayed at the bottom of screen 4
Move and click the left or right button of the mouse 23 to change the shape of the cursor CS from "○" to "△".
Change to. When the left button of the mouse 23 is clicked when the shape of the cursor CS is “Δ”, information that the word delimiter at the position where the cursor CS exists is correct is sent to the analysis result selection unit 8. On the other hand, if the right button is clicked at this time, information that the word division at the position where the cursor CS exists is incorrect is sent to the analysis result selection unit.

Further, when the cursor CS is located at the place where "Next solution" is displayed at the lower right of the screen of the display 24, when the left or right button of the mouse 23 is clicked,
A signal for selecting the next candidate is sent to the analysis result selection unit 8, and
When the left or right button of the mouse 23 is clicked when the cursor CS is located at the position indicated by "previous solution" in the lower right of the screen, a signal for selecting the previous candidate is sent to the analysis result selection unit 8.

If the left or right button of the mouse 23 is clicked when the cursor CS is located at the position "next sentence" at the bottom right of the screen, information that the currently presented solution is correct is displayed. It is sent to the analysis result selection unit 8.

Further, if the left or right button of the mouse 23 is clicked on the place where "automatic" is displayed at the lower left of the screen, an automatic selection signal is sent to the analysis result selection section, and when it is clicked again, automatic cancellation is performed. Is sent.

Further, when the cursor CS is located at the position indicated as "END" at the bottom right of the screen, clicking the left or right button of the mouse 23 terminates all the operations.

In this way, a plurality of possible solutions for the character string analysis result are sequentially presented on the display according to the user's operation, and the user operates the analysis result selection section to select the analysis result. By doing so, even when it is necessary to generate prosodic symbols based on a large number of documents, the correct answer can be obtained by a simple operation, and a large number of documents can be processed by a simple operation.

[0032]

As described above, according to the present invention,
The analysis result obtained by performing the morphological analysis of the character string is attached to the likelihood and stored in the analysis result storage means, and the analysis result stored in the analysis result storage means is selected by referring to the likelihood by the analysis result selection means. Since the prosodic symbol is generated based on the selected analysis result, even a large amount of documents can be processed quickly and accurately by a simple operation.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a prosody symbol generation apparatus according to the present invention.

FIG. 2 is a diagram illustrating a configuration example of an analysis result selection unit.

FIG. 3 is a diagram showing an example of a morpheme lattice.

FIG. 4 is a diagram showing a specific system configuration example of a prosody symbol generation device according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 document reading unit 2 morpheme dictionary 3 morpheme dictionary lookup unit 4 connection table 5 connection verification unit 6 solution generation unit 7 analysis result storage unit 8 analysis result selection unit 9 prosody generation unit 10 speech synthesis unit 21 control device 22 storage device 23 mouse 24 display

Claims (6)

[Claims] 1. A morpheme analysis means for performing a morpheme analysis of a character string, an analysis result storage means for storing a character string analysis result with a likelihood, and an analysis result stored in the analysis result storage means. A prosody symbol generation device comprising: an analysis result selection means for selecting with reference to the likelihood, and generating a prosody symbol based on the selected analysis result. 2. The analysis result selection means is a selection instruction means for causing the user to give various selection instructions, and a word designation for allowing the user to designate a predetermined word in the presented analysis results. And at least one of a means and a word break position designating means for allowing a user to specify a predetermined word break position in the presented analysis result. The prosody symbol generation device according to claim 1. 3. The selection instruction means has an automatic reading selection unit and / or a next sentence selecting unit, and when the automatic reading selection unit gives an automatic reading selection instruction,
The analysis result selection unit is configured to select, as the correct answer, one of the analysis results stored in the analysis result storage unit that has the highest likelihood, and to process the next character string. If the automatic reading selection unit is not provided, or if the automatic reading selection instruction is not issued even if it is provided, the analysis result selection unit is instructed to select the next sentence by the next sentence selection unit. 3. The prosody symbol generation device according to claim 2, wherein the analysis result currently selected is determined to be the correct answer and the next character string is read. 4. The selection instructing means has a next candidate selecting section and a previous candidate selecting section, and when the next candidate selecting section issues an instruction to select the next candidate, the analysis result selecting section determines the analysis result. Among the analysis results stored in the storage means, the analysis result of the likelihood next to the currently selected analysis result is presented, and when the previous candidate selection unit instructs the selection of the previous candidate, the analysis result is 3. The prosody symbol generation device according to claim 2, wherein the selection means is adapted to present again the analysis result presented before the currently selected analysis result. 5. The word designating unit is a false word designating unit that allows a user to designate a wrong word in the presented analysis results, and / or a correct word designating unit that allows the user to designate a correct word. When the erroneous word is designated by the erroneous word designating section, the analysis result selecting means deletes the analysis result including the erroneous word from the analysis result storing means, and When an analysis result to be presented next is selected from the inside, and when a correct word is specified by the correct word specifying unit, the analysis result selection unit deletes the analysis result that does not include the correct word from the analysis result storage unit. 3. The prosody symbol generation device according to claim 2, wherein the analysis result to be presented next is selected from the remaining analysis results. 6. The word delimiter position designating unit, which allows a user to designate a correct word delimiter position in the presented analysis result, and / or an erroneous word delimiter position. When the correct word segmentation position designating section designates a correct word segmentation position, the analysis result selecting means has no word segmentation at the designated position. When the analysis result is deleted from the analysis result storage means, the analysis result to be presented next is selected from the remaining analysis results, and when the position of the incorrect word break is specified by the incorrect word break position specifying unit. The analysis result selection means deletes the analysis result having a word break at the specified position from the analysis result storage means and selects the analysis result to be presented next from the remaining analysis results. The prosody symbol generation device according to claim 2, wherein