JPH11167398A - Voice synthesizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To eliminate the needs for key operation by a user, and also to make it possible for a voce synthesizer to output voices of only points of headlines or paragraphs of a document, when skip reading it in a voice synthesizer converting character information into voice for outputting. SOLUTION: A voice synthesizer is provided with a format conversion part 10 which analyzes a construction of an input document split into plural sentences, and creates and outputs a format-converted document in which construction information permitting to discriminate the headlines and paragraphs are added to the headlines and the following paragraphs in the input document, and a selecting part 20 which selects predetermined character information in the format-converted document, and creates and outputs the final format- converted document in which the selected information is added to the above format-converted document. And, this voice synthesizer is constituted of a read sentence creating part 30 which creates and outputs character information of the sentence to which the selection information is added from the final format-converted document, and a voice synthesis part 40 converting the read sentence into speech for outputting.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizing apparatus for converting document information into speech and outputting the speech, and more particularly to a technique for skipping and oblique reading which converts only a part of a document into speech and outputs the speech. .

[0002]

2. Description of the Related Art FIG. 31 shows a flow chart of a conventional speech synthesizer having a so-called oblique reading or skipping function (JP-A-7-114537). This speech synthesizer is capable of skipping while a voice is being output if the user presses a skip key for instructing skipping. Hereinafter, FIG.
1 will be described.

[0003] First, in step S1, the voice synthesizing apparatus reads characters of a character string. In step S2, it is determined whether or not the user has pressed the skip key. If the user has not pressed the skip key, the process proceeds to step S3, where the character string is analyzed. In step S4, a synthesized voice corresponding to the analysis result is output. . Thereafter, the process returns to step S1, and the process is repeated. Then, when the user presses the skip key while the voice is being output from the voice synthesizer (step S2), the voice synthesizer proceeds to step S5.

In step S5, it is determined whether or not the currently read character is a delimiter. If the delimiter is a delimiter, the process returns to step S1. If not, the process proceeds to step S6.
Proceed to. Here, the delimiter character is a character such as ".. ,," set by the user in advance. By the loop of step S5 and step S6, skipping is performed until a delimiter appears in the read character.
When a delimiter appears, the process returns to step S1, and converts the character string into a voice again from the character following the delimiter and outputs it.

[0005]

In the conventional speech synthesizer, the user has to operate the skip key each time the reading is skipped, and the hand cannot be released from the step key. In addition, in order to quickly grasp the outline of the document, it was not possible to output only the heading items such as chapters and sections of the document and the essential points of each paragraph by voice.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a voice synthesizing apparatus which does not require a user's key operation for skipping, and a voice synthesizing apparatus which outputs only a heading item of a document or a key point of each paragraph. And

[0007]

According to a first aspect of the present invention, a structure of an input document divided into a plurality of sentences is analyzed, and each sentence of a heading item in the input document and a paragraph following the heading item is formed. A format conversion unit that generates and outputs a format conversion document including character information to be converted and structural information that can identify these heading items and paragraphs; and a predetermined sentence in the format conversion document is selected. A selecting unit that generates and outputs a final format conversion document added to the conversion document; a reading-sentence generating unit that generates and outputs character information of a sentence to which the selection information is added from the final format conversion document as a reading-sentence; And a speech synthesizer for converting a read-out sentence into speech and outputting the speech.

According to a second aspect of the present invention, a heading item is determined based on the document structure tag from an input document having a document structure tag that allows the heading item and paragraph to be identified, and structure information for identifying the heading item is output. A heading discriminating unit, a paragraph discriminating unit that discriminates a paragraph from the input document based on the document structure tag, and outputs a structure information capable of identifying the paragraph, and extracts the character information from the input document. It is provided with a format conversion document output unit for generating and outputting a format conversion document including character information and structural information for identifying the above-mentioned heading items and paragraphs.

In a third aspect of the present invention, a format representing a heading format is checked from the input document, and a heading format analyzing unit for outputting the format is determined, and a character string representing the heading item is determined based on the format. A heading discriminator that outputs structure information capable of identifying a heading item, a paragraph discriminator that determines a paragraph that starts with indentation in a sentence following the heading item, and outputs structural information that can identify the paragraph; And a format conversion document output section for generating and outputting a format conversion document including the character information therein and the structural information for identifying the heading items and paragraphs.

According to a fourth aspect of the present invention, a heading item up to a predetermined hierarchy is selected from an input document having a plurality of hierarchized heading items, and a final format conversion document obtained by adding this selection information to the format conversion document is provided. It has a selector for generating and outputting.

A fifth aspect of the present invention comprises a heading selecting section for selecting a predetermined heading item in the format conversion document and adding this selection information to the format conversion document, and a lower hierarchy of the selected heading item. A paragraph selecting section for selecting a predetermined paragraph to be added and adding this selection information to the format conversion document, and a paragraph for selecting a predetermined sentence in the selected paragraph and adding this selection information to the format conversion document And a selection unit.

A sixth aspect of the present invention is provided with a sentence selection unit for selecting a head sentence and a tail sentence of the paragraph selected by the paragraph selection unit, and adding the selected information to the format conversion document.

According to a seventh aspect of the present invention, when the number of characters of a plurality of skipped sentences from a position after the selected sentence to a position before the next selected sentence exceeds a predetermined value, an intermediate position between the plurality of skipped sentences is set. Is provided, and a sentence selection unit for selecting this sentence and adding this selection information to the format conversion document is provided.

According to an eighth aspect of the present invention, a sentence boundary keyword indicating a break of a series of sentences is detected from a plurality of sentences constituting the paragraph selected by the paragraph selection section, and the position of the sentence boundary keyword is output. A boundary keyword detecting unit; and a paragraph adding unit that divides the paragraph at the position of the detected sentence boundary keyword and adds a new paragraph. The sentence selecting unit includes a new sentence added by the paragraph adding unit. Select the appropriate paragraph,
This selection information is added to the format conversion document.

A ninth aspect of the present invention is provided with a keyword extracting section for examining the appearance frequency of an independent word from the sentence of the paragraph to which the selection information is added, and outputting the independent word having a high appearance frequency as a read-out sentence.

According to a tenth aspect of the present invention, there is provided a modifier processing unit for detecting a continuation modifier from the sentence of the paragraph to which the selection information is added, and outputting a sentence in which the continuation modifier is deleted as a read-out sentence. is there.

According to an eleventh aspect, when the number of characters of a sentence to be skipped after the sentence to which the selection information is added and before the sentence to which the selection information is added next exceeds a predetermined value, the reading is performed. The present invention further comprises a skipped character number notifying unit for adding a skipped notification sentence for notifying the number of characters of the sentence to the read-out sentence.

A twelfth invention is a control section which outputs character information in a speech sentence output from the speech sentence generation section as a speech string, and further issues a command to generate a signal sound in accordance with a boundary of each sentence in the speech sentence. A text-to-speech conversion unit that converts the read-out character string output from the control unit to a synthesized voice, a signal sound generation unit that outputs a signal sound in accordance with the generation command, and adds the synthesized sound and the signal sound. And an adder for generating and outputting an output sound.

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 Embodiment 1
In order to know the outline of a structured document such as an HTML document, only the header and heading of the document are converted into audio and output. FIG. 1 is an overall configuration diagram of the speech synthesis apparatus according to the first embodiment, FIG. 2 is a configuration diagram of a format conversion unit 10, and FIG.
FIG. 3 is a configuration diagram of the speech synthesis unit 40. The input document 50, the first format conversion document 60, the final format conversion document 70, and the read-out sentence 80, which are the respective components of FIG.
As shown in FIG. Note that the input document 5 in the first embodiment
0 is HT including a document structure tag indicating the structure of the document
An input document 50 is a structured document such as an ML document.
The description is made assuming that the document is an ML document. First, format converter 1
0 converts the input document 50 into a format that can be handled by the selection unit 20 and the read-aloud sentence generation unit 30 that follow the format conversion unit 10. Hereinafter, the details of the processing will be described.

In an HTML document, ■ <HEAD> ■, ■ <H1> are used to indicate the header, heading, figure, table, and paragraph in the text of the document.
Document structure tags such as ■, ■ <H2> ■, ■ <IMG> ■, ■ <P> ■ are used. Further, the heading has a hierarchical structure represented by document structure tags of {<H1>}, {<H2>}, ...., {<H6>}. The heading determination unit 11
Based on such a document structure tag, shown in FIG.
A header item such as a header or a heading of the input document 50 in the HTML format is determined, and positional information and hierarchical information of the heading item in the input document 50 are output to the paragraph determining unit 12. The hierarchy information output by the heading discrimination unit 11 is a number representing the hierarchy of the headings of the document structure tags {<H1>}, {<H2>},..., {<H6>}. However, it is assumed that the header of the document is at the highest level higher than the heading.

The paragraph discriminating unit 12 discriminates a paragraph boundary of a part which is not a heading item of the input document based on a document structure tag {<P>} indicating a paragraph boundary, and determines a paragraph boundary such as a line number in the input document. The position information 52 is output to the format conversion document output unit 13. The format conversion document output unit 13 generates character information of the first format conversion document 60 by deleting the figure and the document structure tag from the input document 50,
Further, as shown in FIG. 5, on the left side of a heading item such as a document title or heading of the character information of the first format conversion document 60, the hierarchy information of the symbols {H0} and {H1} is shown. Is added to the left of the first line of the paragraph of the character information of the first format conversion document 60 to indicate the beginning of the paragraph.
(1) The first format conversion document 60 is generated and output by adding the position information of the symbol P ■. The symbol ■ H0 ■ indicates the document header, ■ H1 ■, ■ H2 ■, ...., ■ H6
■ indicates the heading as in the HTML document, and {P} indicates the beginning of the paragraph.

Next, the selector 2 following the format converter 10
0 will be described. The selection unit 20 outputs the first format conversion document 6
Based on the hierarchical information of the heading item described in 0,
From the first format conversion document 60, only the heading items up to the upper K stages are selected, and the selected information is added to the first format conversion document 60, whereby the final format conversion shown in FIG. A document 70 is generated and output. Where K
Is an integer set in advance, and is 2 in the present embodiment. The selection information is {H0}, {H2} in FIG.
■, {1} on the right side of {H3}, where {1} indicates selection, and those without any number indicate no selection.

Next, the text-to-speech generation unit 30 subsequent to the selection unit 20 will be described. The reading sentence generation unit 30
Using the final format conversion document 70, a read-aloud sentence 80 to be input to the speech synthesis unit 40 is generated and output.
The text-to-speech generation unit 30 generates, as a text-to-speech, character information of only the heading item specified by the selection information from the final format conversion document 70. Further, the reading-sentence generating unit 30 adds boundary information indicating that the information is character information for each heading item to the character information of the reading-sentence,
The read-out sentence 80 shown in FIG. 7 is generated and output. The boundary information is a symbol {T} at the left end in FIG.

Finally, the speech synthesizing section 40 subsequent to the reading sentence generating section 30 will be described with reference to FIG. First, when detecting the boundary information in the read-aloud sentence 80, the control unit 41
A signal sound generation command 82 is given to the signal sound generation unit 43. When receiving the generation command 82, the signal sound generation unit 43 receives the signal sound 92.
Is output to the adder 44. After the signal sound 92 is output, the control unit 41 sends the text information of the first heading item of the read-out sentence 80 to the text-to-speech conversion unit 42 in the read-out character string 8.
Send as 1. The text-to-speech converter 42 converts the input text information of the heading item into a synthesized speech 91 and outputs the synthesized speech 91 to the adder 44. The adder 44 adds the synthetic sound 91 to the signal sound 92 and outputs an output sound 90. Hereinafter, the control unit 41, the signal sound generation unit 43, and the text-to-speech conversion unit 42
Performs the same processing for the second and subsequent heading items. As a result, the heading items of the hierarchies up to the upper K stages are converted from the voice synthesizing unit 40 into output voices 90 accompanied by a signal sound before each heading item and output.

As described above, in the present embodiment,
The user does not need to perform key operations such as the skip key,
Only the headline items of the input document 50 can be heard as voice, and the contents of the input document 50 can be grasped in a short time.

Embodiment 2 FIG. In the second embodiment, in order to know the outline of the document in more detail than in the first embodiment, in addition to the header and heading of the document, the beginning and end sentences of the paragraph are also converted to speech and output. . The overall configuration diagram of the speech synthesizer of the present embodiment is the same as FIG. The configuration of the format conversion unit 10 of the present embodiment is as shown in FIG. 2, the configuration of the selection unit 20 is as shown in FIG. 8, and the configuration of the speech synthesis unit 40 is as shown in FIG. The first format conversion document 60, the second format conversion document 61, the third format conversion document 62, and the final format conversion document 70, which are input / output documents of the selection unit 20, are respectively shown in FIGS.
FIG. 13 shows a reading sentence 80 which is an output document of the reading sentence generation unit 30. Here, the format conversion unit 10
Has the same configuration and operation as those of the first embodiment, and thus the description is omitted.

Next, the selector 2 following the format converter 10
0 will be described with reference to FIGS. 8 and 9 to 13. In the present embodiment, the selection unit 20 includes a heading selection unit 21, a paragraph selection unit 22, and a sentence selection unit 23, as shown in FIG. The heading selection unit 21 selects a higher-order one of the first format conversion documents 60 based on the position information and the hierarchy information of the heading items described in the first format conversion document 60 shown in FIG. By selecting only the heading items of the hierarchy up to the K stage and adding the selected information to the first format conversion document 60, a second format conversion document 61 shown in FIG. I do. Here, K is an integer set in advance as in the first embodiment, and is 2 in the present embodiment.

The paragraph selecting section 22 stores the second format conversion document 6
1. Based on the selection information of the heading item described in 1 and the position information of the paragraph, a lower hierarchy of the heading item specified by the selection information is constructed from the second format conversion document 61. By selecting a paragraph and adding the selected information to the second format conversion document 61, a third format conversion document 62 shown in FIG. 11 is generated and output to the sentence selection unit 23. In the selection information shown in FIG. 11, {1-6} is the first to sixth sentences in the lower hierarchy of “Heading 1”, “Sentence 1.
・. "To" sentence 6 ... "is one paragraph, and the selection information {7-11} is a sentence from the seventh to eleventh in the lower hierarchy of" heading 1 ". 7 ...
・. ... To “sentence 11...” Indicate one paragraph.

The sentence selecting section 23 outputs the third format conversion document 62
12 is selected as the first sentence and the last M sentence of the paragraph specified by the selection information as the first sentence, and the selected information is added to the third format conversion document 62 in FIG.
And generates and outputs the final format conversion document 70 shown in FIG. Here, L and M are predetermined integers of about 1 to 3. FIG. 12 shows an example in which L is 2 and M is 1, and {1, 2, 6} of the selection information is obtained by using the first and second sentences in the lower hierarchy of “Heading 1” as the first sentence. The fact that the sixth sentence has been selected as the last sentence is indicated by the selection information of {7,8,11}.
Indicates that the seventh and eighth sentences in the lower hierarchy of “Heading 1” are selected as the first sentence and the eleventh sentence is selected as the last sentence.

Next, the text-to-speech generation unit 30 subsequent to the selection unit 20 will be described. The reading-sentence generating unit 30 generates, as a reading-out sentence, character information of a paragraph including only the heading item, the head sentence, and the tail sentence specified by the selection information from the final format conversion document 70. Further, the text-to-speech generation unit 30 adds the heading item in the text information of the text-to-speech and the boundary information between the head sentence and the text to the text information of the text-to-speech, and generates and outputs the text-to-speech 80 shown in FIG. I do.

Finally, the speech synthesizing section 40 following the reading sentence generating section 30 will be described with reference to FIG. Voice synthesis unit 4
0 is the same as that of the first embodiment, but since the content of the input sentence 80 is different,
The operation will be described with the heading item, the head sentence, and the tail sentence of the item referred to as items.

First, when detecting the boundary information in the read-out sentence 80, the control unit 41 gives a signal generation command 82 to the signal generation unit 43. Signaling sound generation unit 43 generates generation command 82
Upon receiving the signal, a signal sound 92 is output to the adder 44. After the signal sound 92 is output, the control unit 41 sends the text information of the first heading item of the read-aloud sentence 80 to the text-to-speech conversion unit 42 as a read-out character string 81. The text-to-speech converter 42 converts the input text information of the heading item into a synthesized speech 91 and outputs the synthesized speech 91 to the adder 44. Hereinafter, the control unit 4
1, the signal sound generator 43 and the text-to-speech converter 42
The same processing is performed for the items of the first sentence and the last sentence after the first sentence. As a result, the speech synthesizing unit 40 converts the heading items of the hierarchy up to the upper K stage and the beginning sentence and the ending sentence of the paragraph of the lower hierarchy into output speech 90 with a signal sound before each. Is output.

As described above, in the present embodiment,
The user does not need to perform key operations such as the skip key,
It is possible to listen to the heading item of the hierarchy up to the upper K stages and the beginning and end sentences of the paragraphs in the lower hierarchy as speech, and to grasp the contents of the input document 50 in a short time.

Embodiment 3 According to the third embodiment, in order to know the outline of a document which does not have a special tag for structuring and is composed of only character information, a hierarchy such as a heading or a paragraph in the input document 50 based on a predetermined criterion is used. Determine the structure,
The heading and the sentence at the beginning and end of the paragraph are converted into voice and output, and the format conversion unit 10 has the feature.

The overall configuration of the speech synthesizing apparatus according to this embodiment is the same as that shown in FIG. Format conversion unit 10 of the present embodiment
14 is as shown in FIG. 14, the configuration of the selector 20 is as shown in FIG. 8, and the configuration of the speech synthesizer 40 is as shown in FIG. Format conversion unit 1
FIG. 15 and FIG. 16 show an input document 50 and a first format conversion document 60 which are input / output documents of No. 0, respectively. Here, the selection unit 20, the reading sentence generation unit 30, and the speech synthesis unit 40
Since the configuration and operation are the same as those of the second embodiment, the description will be omitted, and only the format conversion unit 10 will be described.

As shown in FIG. 14, the format conversion unit 10
It is composed of a heading format analyzing unit 14, a heading determining unit 11, a paragraph determining unit 12, and a format conversion document output unit 13. First, the heading format analysis unit 14 analyzes the heading format from the whole or a part of the input document 50. Specifically, the format of the headline of the document is "Chapter 1, Chapter 2, Chapter 3, ...
.. "or" 1, 2., 3., ...
・ ・ ”, Etc., and check heading format 5
3 is output.

Next, the heading discriminating unit 11 uses the heading format 53 given by the heading format analyzing unit 14 to discriminate the heading items in the input document 50. In the example shown in FIG. 15, the character string “1, 2., 3,...” At the head of the line is determined as a heading item. Then, this character string, that is, the line number and hierarchy in the input document of the heading item,
Output as position information and hierarchical information of the heading item.

The paragraph determining section 12 uses the hierarchical information of the heading item provided by the heading determining section 11 to input the input document 5.
The paragraph boundary of the part other than the heading item of 0 is determined. Specifically, as shown in FIG. 15, in the input document 50,
The line indented at a place other than the heading item is the head of the paragraph. Then, the line number of the paragraph boundary in the input document is output as position information. As shown in FIG.
The format conversion document output unit 13 adds, to the left side of a heading item such as a document title or a heading of the input document 50, hierarchical information of symbols {H0} and {H1} indicating the hierarchy of the heading item. To the left of the first line of the 50 paragraphs,
The character information of the first format conversion document 60 is generated and output by adding the position information of the symbol {P} indicating the beginning of the paragraph.

The first format conversion document 60 generated as described above has the same format as the first format conversion document 60 input to the selection unit 20 of the second embodiment. Therefore, the selecting unit 20, the reading-sentence generating unit 30, and the speech synthesizing unit 40 subsequent to the format converting unit 10 according to the present embodiment perform the same operations as described in the second embodiment. Therefore, in the speech synthesizer according to the present embodiment, even when the input document 50 is composed of only character information having no special tag for structuring, the user does not perform a key operation such as a skip key. It is possible to listen to the heading item of the hierarchy up to the upper K stages and the beginning and end sentences of the paragraphs in the lower hierarchy as speech, and to grasp the contents of the input document 50 in a short time.

Embodiment 4 FIG. The fourth embodiment is to prevent the number of characters to be skipped from the first sentence to the last sentence from being too large in the second embodiment. The overall configuration diagram of the speech synthesizer of the present embodiment is the same as FIG. Format conversion unit 1 of the present embodiment
2, the configuration of the selector 20 is as shown in FIG. 17, and the configuration of the speech synthesizer 40 is as shown in FIG. The fourth format conversion document 63 and the final format conversion document 70, which are part of the input / output document in the selection unit 20, are respectively shown in FIGS. FIG. 20 shows a reading sentence 80 as a document. Here, the format conversion unit 10 has the same configuration and operation as those of the first embodiment, and thus the description is omitted.

As shown in FIG. 17, the selection section 20 includes a heading selection section 21, a paragraph selection section 22, a sentence selection section 23, and a skip restriction section 24. The selecting unit 20 according to the present embodiment is the same as the second embodiment except that the skip restricting unit 24 is added to the selecting unit 20 according to the second embodiment. explain.

The skipping restriction unit 24 inputs the fourth format conversion document 63 shown in FIG. 18 and counts the number of characters to be skipped from after the first sentence to before the last sentence. And
When the number of characters to be skipped exceeds a predetermined value and extends to a plurality of sentences, a middle sentence of the plurality of skipped sentences is selected as an intermediate sentence. FIG. 19 shows the final format conversion document 70 which is the output of the skipping restriction unit 24. The skipping of the first paragraph of the lower hierarchy of “Heading 1” and the first paragraph of the lower hierarchy of “Heading 2” are performed. Since the number of characters exceeds a predetermined value, “sentence 4...” Which is the fourth sentence of the lower hierarchy of “heading 1” and “sentence 4” which is the fourth sentence of the lower hierarchy of “heading 2” ... Have been selected as intermediate sentences.

The reading sentence generating section 30 generates the reading sentence 80 in FIG. 20 from the final format conversion document 70 in FIG. This figure shows that the intermediate sentence selected by the selection unit 20 is added with boundary information and is written out in the read-aloud sentence 80.

Finally, the speech synthesizer 40 will be described.
The configuration of the voice synthesizer 40 is the same as that described in the second embodiment. The difference from the description of the second embodiment is that the read-aloud sentence 80, which is the input of the speech synthesizer 40, is different from FIG. 13 of the second embodiment in FIG.
The difference is that an intermediate sentence is added to the lines 11 and 12. Therefore, in the second embodiment, regardless of the number of characters to be skipped from after the initial sentence to before the final sentence, only the heading item of the hierarchy up to the upper K-level and the opening sentence and the final sentence of the paragraph of the lower hierarchy are reduced. , The output sound 90 with a signal sound before each
In contrast, in the present embodiment,
If the number of characters to be skipped after the beginning sentence to before the end sentence is large, in addition to the heading items of the hierarchy up to the upper K level and the beginning and end sentences of the paragraphs in the lower hierarchy, the intermediate sentence It is converted into an output sound 90 accompanied by a signal sound and output.

As described above, in the present embodiment,
The user does not need to perform key operations such as the skip key,
It is possible to listen to the heading item and the first sentence and the last sentence of the heading items of the hierarchy up to the upper K stage and the paragraphs of the lower hierarchy.
If the number of characters to be skipped after the initial sentence to before the final sentence exceeds a predetermined value, the intermediate sentence between the initial sentence and the final sentence can be heard as a voice. Can be grasped in a short time, and even in a large paragraph having many characters.

Embodiment 5 FIG. In the fifth embodiment, a new paragraph boundary is added based on a sentence boundary keyword indicating a unitary break of a series of sentences. Sentence boundary keywords are "first,""first,""second," etc.
It is a collocation that mainly represents a change in topic. The overall configuration diagram of the speech synthesizer of the present embodiment is the same as FIG. The configuration of the format conversion unit 10 of the present embodiment is as shown in FIG. 2, the configuration of the selection unit 20 is as shown in FIG. 21, and the configuration of the speech synthesis unit 40 is as shown in FIG. FIG. 22 shows a third format conversion document 62, a fourth format conversion document 63, and a final format conversion document 70, which are input / output documents of the selection unit 20. Here, the format conversion unit 10
Since the configuration and the operation are the same as those of the first embodiment, the description is omitted.

Next, the selection section 20 will be described. As shown in FIG. 21, the selection unit 20 includes a heading selection unit 21, a paragraph selection unit 22, a sentence boundary keyword detection unit 25, and a paragraph addition unit 2.
6. Sentence selection unit 23. The selection unit 20 of the present embodiment is the same as the second embodiment except that a sentence boundary keyword detection unit 25 and a paragraph addition unit 26 are added to the selection unit 20 of the second embodiment. Only the boundary keyword detecting section 25 and the paragraph adding section 26 will be described.

The sentence boundary keyword detecting unit 25 is configured as shown in FIG.
The third format conversion document 62 shown in (a) is input, and the presence or absence of a sentence boundary keyword is checked at the beginning of each sentence of the lower hierarchy of the heading item whose selection information is {1}. If detected, the position information 64 of the sentence boundary keyword in the third format conversion document 62 is output. If the position of the sentence boundary keyword does not overlap with the paragraph boundary indicated by the selection information of the third format conversion document 62, the paragraph adding unit 26 adds the position of the sentence boundary keyword as a new paragraph boundary, and The fourth shown in FIG.
Is output. In FIG. 22B, since a sentence boundary keyword is at the beginning of “sentence 4.
This indicates that one paragraph in the lower hierarchy of “Heading 1” has been divided into two paragraphs.

The sentence selection unit 23 sets the first L sentence of the paragraph specified by the selection information of the fourth format conversion document 63 as the first sentence and the last M sentence as the last sentence for each of the divided paragraphs. Selected as a sentence, and the selected information is sent to the fourth format conversion document 6
3 to generate and output the final format conversion document 70 shown in FIG. In the present embodiment, L
And M are both 1. The operation of the text-to-speech generation unit 30 and the speech synthesis unit 40 following the selection unit 20 is described in the second embodiment.
Therefore, the description is omitted.

As described above, in the selecting section 20 of the present embodiment, when the sentence boundary keyword detecting section 25 detects a sentence boundary keyword at the beginning of a sentence in a paragraph, the paragraph adding section 2
6 divides a paragraph, and the sentence selection unit 23 selects a head sentence and a tail sentence from each of the divided paragraphs.

Therefore, the structure tag representing the paragraph of the input document in the HTML format, or the indentation representing the beginning of the paragraph of the input document having no structure tag may be a position indicating a meaningful unit of the sentence or a position at which the topic changes. If the input document 50 is large, the sentence boundary keyword can be used as a clue to divide the paragraph, and the sentence in each of the divided paragraphs can be converted into output speech 90 and output. , The user can grasp the contents of the document accurately and in a short time.

Embodiment 6 FIG. In the sixth embodiment, an independent word having a high appearance frequency in a paragraph is used as a keyword, and only the keyword is read aloud instead of reading out a sentence in the paragraph. The overall configuration diagram of the speech synthesizer of the present embodiment is the same as FIG. Format conversion unit 1 of the present embodiment
2, the configuration of the selection unit 20 is as shown in FIG. 23, the configuration of the read-aloud sentence generation unit 30 is as shown in FIG. 24, and the configuration of the speech synthesis unit 40 is as shown in FIG. A final format conversion document 70, which is an input / output document of the reading sentence generating unit 30, and a reading sentence 80
Is shown in FIG. Here, the format conversion unit 10 has the same configuration and operation as those of the first embodiment, and thus the description is omitted.

Next, the selection section 20 will be described. As shown in FIG. 23, the selection unit 20 includes a heading selection unit 21 and a paragraph selection unit 22. This configuration is obtained by removing the sentence selection unit 23 from the selection unit 20 of the second embodiment shown in FIG. Heading selection section 21 and paragraph selection section 22
Is the same as that in the second embodiment except that the output name of the paragraph selecting unit 22 is the third format conversion document 62 in the second embodiment, but is the final format conversion document 70 in the present embodiment. 2, the description of the operation is omitted.

FIG. 25A shows a final format conversion document 70 output from the selection unit 20. This figure shows only one heading and one paragraph below the heading to show the gist of the present embodiment. At the left end of the first line, a symbol {H2} indicating that this line is a heading item is attached, and at the right thereof, a symbol {1} indicating that the heading item is selected. At the left end of the second line is a symbol {P} indicating that it is a paragraph, and to the right is {1} indicating that the first to sixth sentences in the lower hierarchy of the heading item are the range of the paragraph. The symbol of -6 ■ is attached.

Next, the text-to-speech generating unit 30 will be described. As shown in FIG. 24, the reading-sentence generating unit 30
It comprises a keyword extracting unit 32 and a reading sentence output unit 31. First, the keyword extracting unit 32 receives the final format conversion document 70 described above as an input, and checks the frequency of independent words in the sentence of the paragraph indicated by the selection information such as {1-6}. Then, the independence word having a high frequency is extracted and output as the keyword 71 of the paragraph. Next, the reading sentence output unit 31
The character information of the text to be read, which includes only the heading item specified by the selection information and the keyword 71 output from the keyword extracting unit 32, is generated. Further, the read-sentence generating unit 30 adds the boundary information of the heading item in the character information of the read-out sentence and the boundary information of the paragraph to which the extracted keyword 71 belongs to the character information of the read-out sentence. The read-out sentence 80 shown in (b) is generated and output. Note that, as in the other embodiments, the symbol {T} at the left end of the figure is the boundary information.

Finally, the speech synthesizer 40 will be described.
The configuration of the voice synthesizer 40 is the same as that described in the second embodiment. The difference from the description of the second embodiment is that in the read-aloud sentence 80 that is the input of the speech synthesis unit 40, the part that is the beginning sentence and the end sentence is a keyword in the present embodiment. Therefore, in the second embodiment, the heading item of the hierarchy up to the upper K-level and the beginning sentence and the end sentence of the paragraph of the lower hierarchy are converted into the output sound 90 accompanied by a signal sound before each and output. On the other hand, in the present embodiment, the heading item of the hierarchy up to the K-th level and the keyword of the paragraph in the lower hierarchy are converted into the output sound 90 accompanied by a signal sound before each and output.

As described above, in the present embodiment,
The user does not need to perform key operations such as the skip key,
Because you can listen to the keyword of the heading item of the hierarchy up to the upper K stage and the paragraph of the lower hierarchy,
The contents of the input document 50 can be grasped in a short time.

Embodiment 7 FIG. In the seventh embodiment, a sentence in a paragraph is not read aloud as it is, but a continuous modification clause of the sentence, that is, a phrase that modifies a nominative is deleted and read. The overall configuration diagram of the speech synthesizer of the present embodiment is the same as FIG. Format conversion unit 10 of the present embodiment
2 is shown in FIG. 2, the configuration of the selection section 20 is shown in FIG. 23, the configuration of the reading-sentence generation section 30 is shown in FIG. 26, and the configuration of the speech synthesis section 40 is shown in FIG. FIG. 27 shows the final format conversion document 70 and the read-out sentence 80, which are input / output documents of the read-out sentence generation unit 30. Here, the format conversion unit 10 has the same configuration and operation as those of the first embodiment, and thus the description is omitted. In addition, the selection unit 20 has the same configuration and operation as those of the sixth embodiment, and thus the description thereof is omitted.

Next, the reading sentence generating unit 30 will be described. As shown in FIG. 26, the text-to-speech generating unit 30 includes a modifier processing unit 33 and a text-to-speech output unit 31. First, the modifier processing unit 33 receives the final format conversion document 70 shown in FIG. 27 (a) as an input, and selects between paragraphs in a sentence indicated as a paragraph by selection information such as {1-4}. Analyze the modification relations and detect adnominal modification clauses that modify the nominal. Then, the sentence 72 from which the adnominal modifier clause is deleted is output.

Next, the read-out sentence output unit 31 outputs the character information of the read-out sentence composed only of the heading item designated by the selection information and the sentence 72 in which the adjunct modifier output from the modifier processing unit 33 is deleted. Generate Further, the reading-sentence generating unit 30 adds the boundary information of the heading item in the character information of the reading-out sentence and the boundary information of the paragraph to which the sentence from which the modifier clause has been deleted to the character information of the reading-out sentence, and adds the reading-out sentence 80.
Generate and output FIG. 27 (b) shows a read-aloud sentence 80, where the symbol {T} at the left end is the boundary information, and the right side is the character information of the sentence 72 from which the adnominal modifier clause has been deleted.

Finally, the speech synthesizer 40 will be described.
The configuration of the voice synthesizer 40 is the same as that described in the second embodiment. The difference from the description of the second embodiment is that, in the read-aloud sentence 80 which is the input of the speech synthesis unit 40, the first sentence and the last sentence of the second embodiment are replaced by the sentence from which the modified clause is deleted in the present embodiment. It is a point that has become. Therefore, in the second embodiment, the heading items of the hierarchies up to the K-th level and the beginning and end sentences of the paragraphs in the lower hierarchy are converted into output sound 90 accompanied by a signal sound before each and output. On the other hand, in the present embodiment, the sentence in which the heading item of the hierarchy up to the upper K level and the union modifier in the paragraph of the lower hierarchy are deleted is output sound 9 with a signal sound before each sentence.
It is converted to 0 and output.

As described above, in the present embodiment,
The user does not need to perform key operations such as the skip key,
Since the sentence in which the heading item of the hierarchy up to the upper K stage and the adjunct modifier in the lower hierarchy can be heard as a voice, the contents of the input document 50 can be grasped in a short time.

Embodiment 8 FIG. In the eighth embodiment, the number of characters to be skipped when reading the first sentence and the last sentence in a paragraph is notified by voice. The overall configuration diagram of the speech synthesizer of the present embodiment is the same as FIG. The configuration of the format conversion unit 10 according to the present embodiment is as shown in FIG. 2, the configuration of the selection unit 20 is as shown in FIG. 8, the configuration of the reading sentence generation unit 30 is as shown in FIG. 28, and the configuration of the speech synthesis unit 40 is as shown in FIG. Further, the final format conversion document 7 which is an input / output document of the reading-sentence generation unit 30
0 and the sentence 80 are shown in FIGS. 29 and 30, respectively.
Here, the format conversion unit 10 has the same configuration and operation as those of the first embodiment, and thus the description is omitted. The selection unit 20
Has the same configuration and operation as those of the second embodiment, so that the description thereof will be omitted.

Next, the reading sentence generating unit 30 will be described. As shown in FIG. 28, the reading-sentence generating unit 30
It comprises a skipped character number notification unit 34 and a read-out sentence output unit 31. First, the skipped-character-number notifying unit 34 counts the number of skipped characters from after the first sentence to before the last sentence of the final format conversion document 70 shown in FIG. Then, when the number of characters to be skipped exceeds a predetermined value, a notification message for notifying the number of characters to be skipped is created. The format of the skip notification message 73 is, for example, "82 characters are skipped."

Next, the read-sentence output unit 31 outputs
In the heading item selected in 0 and the head and tail sentences,
Furthermore, the skip notification message 73 output from the skip character notification unit 34 is inserted to generate character information of the text to be read. In addition, the reading sentence output unit 31 generates and outputs a reading sentence 80 shown in FIG. 30 by adding boundary information to the character information of the reading sentence.

Finally, the speech synthesizer 40 will be described.
The configuration of the voice synthesizer 40 is the same as that described in the second embodiment. The difference from the description of the second embodiment is that the text-to-speech sentence 80 input to the speech synthesis unit 40 has a skipped notification text 73 added to the fourth and eleventh lines as shown in FIG.

Therefore, in the second embodiment, regardless of the number of characters to be skipped from the beginning sentence to the end sentence, regardless of the number of characters skipped, the first sentence and the last sentence of the upper-level K-level heading items and the lower-level paragraphs Only the sentence is output after being converted into an output sound 90 accompanied by a signal sound before each sentence. On the other hand, in the present embodiment, when the number of characters to be skipped from the beginning sentence to the end of the last sentence is large, the heading items of the hierarchy up to the upper K level and the beginning of the paragraph of the lower hierarchy In addition to the sentence and the tail sentence, the skipped notice sentence is converted into an output voice 90 accompanied by a signal sound and output.

As described above, in the present embodiment,
The user does not need to perform key operations such as the skip key,
It is possible to listen to the heading item and the first sentence and the last sentence of the heading items of the hierarchy up to the upper K stage and the paragraphs of the lower hierarchy.
Furthermore, when the number of characters to be skipped from after the initial sentence to before the last sentence exceeds a predetermined value, the number of characters can be heard as a voice, so that the contents of the input document 50 can be read in a short time. It is possible to grasp, and furthermore, it is possible to know the location and the number of skipped characters that may not be accurately grasped due to the large number of skipped characters, by the skipped notification message.

[0069]

Since the present invention is configured as described above, it has the following effects.

In the first invention, from an input document divided into a plurality of sentences, character information of each of a heading item and a paragraph, structural information that can identify them, and selection information for selecting predetermined character information are obtained. Is generated, and predetermined character information is generated as a text-to-speech based on the format conversion document and is output as voice. Only the part can be listened to as sound, and the contents of the input document can be grasped in a short time.

In the second invention, the heading item and the paragraph are distinguished from the input document having the document structure tag in which the heading item and the paragraph can be identified. Therefore, the input document has a structure tag like an HTML document. Only a part of the structured document can be heard as voice, and the contents of the input document can be grasped in a short time.

In the third invention, a format representing a heading format is checked from an input document, and a character string representing a heading item and a paragraph is determined based on this format. Only a part of the document to which no is given can be heard as voice, and the contents of the input document can be grasped in a short time.

In the fourth invention, a heading item of a predetermined hierarchy is selected from an input document having a plurality of hierarchized heading items. Therefore, only the heading items from the top of the input document to the desired hierarchy are selected. Can be heard as voice, and the contents of the input document can be grasped in a short time.

In the fifth invention, a heading item and a paragraph are selected from the format conversion document of the input document, and a predetermined sentence in this paragraph is further selected. It is possible to listen to the heading item and a part of the sentence of the paragraph in the lower hierarchy as voice, and to grasp the main points of the input document in a short time.

In the sixth invention, a heading item and a paragraph are selected from the input document, and a head sentence and a tail sentence of this paragraph are further selected. Only the first sentence and the last sentence of the lower hierarchical paragraph can be listened to as speech, and the gist of the input document can be grasped in a short time.

According to the seventh invention, when the number of characters of a plurality of skipped sentences from after the selected sentence to before the next selected sentence exceeds a predetermined value, the number of characters in the plurality of skipped sentences is set in the middle. Since the selected sentence is selected, if the number of characters skipped in a paragraph exceeds a predetermined value,
The middle sentence of the skipped sentence can be heard as speech, and the contents of the input document can be grasped in a short time, and even in a large paragraph.

In the eighth invention, a sentence boundary keyword indicating a break of a series of sentences is detected from a plurality of sentences constituting the selected paragraph, and the paragraph is divided at the position of the sentence boundary keyword to form a new paragraph. Because a paragraph is added, a part of a new paragraph that is divided by the sentence boundary keyword at the beginning of the paragraph can be heard as audio, and the contents of the input document can be edited in a short time You can understand the contents even in paragraphs.

In the ninth invention, the appearance frequency of the independent word is checked from the sentence of the paragraph, and the independent word having a high appearance frequency is output as a text-to-speech. As a result, the contents of the input document can be grasped in a short time.

In the tenth aspect, the adnominal modifier clause is detected from the sentence of the paragraph, and the sentence from which the adnominal modifier clause is deleted is output as a read-out sentence. Therefore, the adnominal modifier clause of the selected paragraph is deleted. The sentence can be heard as voice, and the contents of the input document can be grasped in a short time.

In the eleventh invention, when the number of characters of the skipped sentence from after the selected sentence to before the next selected sentence exceeds a predetermined value, the number of characters of the skipped sentence is notified. Therefore, if the number of characters skipped in a paragraph exceeds a predetermined value, the number of characters can be heard as audio, and the content may not be accurately understood due to the large number of skipped characters. Can be known by voice notification of the number of characters.

According to the twelfth invention, when outputting a read-aloud sentence by voice, a signal sound is generated in accordance with the boundary of each sentence in the read-out sentence. Therefore, the boundary of the read-aloud sentence converted into voice is known by the signal sound. be able to.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of a speech synthesis device according to a first embodiment.

FIG. 2 is a configuration diagram of a format conversion unit according to the first embodiment.

FIG. 3 is a configuration diagram of a speech synthesis unit according to the first embodiment.

FIG. 4 is an input document input to the format conversion document according to the first embodiment.

FIG. 5 is a first format conversion document output from the format conversion document according to the first embodiment.

FIG. 6 illustrates a final format conversion document output from the selection unit according to the first embodiment.

FIG. 7 is a reading sentence output from a reading sentence generation unit according to the first embodiment.

FIG. 8 is a configuration diagram of a selection unit according to the second embodiment.

FIG. 9 is a diagram illustrating a first example output from the format conversion unit according to the second embodiment.
Format conversion document.

FIG. 10 illustrates a second format conversion document output from the heading selection unit according to the second embodiment.

FIG. 11 shows a third format conversion document output from the paragraph selection unit according to the second embodiment.

FIG. 12 illustrates a final format conversion document output from the sentence selection unit according to the second embodiment.

FIG. 13 is a reading sentence output from a reading sentence generation unit according to the second embodiment.

FIG. 14 is a configuration diagram of a format conversion unit according to the third embodiment.

FIG. 15 illustrates an input document input to the format conversion unit according to the third embodiment.

FIG. 16 illustrates a first format conversion document output from a format conversion unit according to the third embodiment.

FIG. 17 is a configuration diagram of a selection unit according to the fourth embodiment.

FIG. 18 is a diagram illustrating a fourth example output from the sentence selection unit according to the fourth embodiment.
Format conversion document.

FIG. 19 illustrates a final format conversion document output from the skipping restriction unit according to the fourth embodiment.

FIG. 20 is a reading sentence output from a reading sentence generation unit according to the fourth embodiment.

FIG. 21 is a configuration diagram of a selection unit according to the fifth embodiment.

FIG. 22 shows a third format conversion document, a fourth format conversion document, and a final format conversion document input and output by the selection unit according to the fifth embodiment.

FIG. 23 is a configuration diagram of a selection unit according to the sixth embodiment.

FIG. 24 is a configuration diagram of a reading sentence generation unit according to the sixth embodiment.

FIG. 25 illustrates a final format conversion document and a read-aloud sentence input / output by the selection unit according to the sixth embodiment.

FIG. 26 is a configuration diagram of a reading sentence generation unit according to the seventh embodiment.

FIG. 27 shows a final format conversion document and a read-aloud sentence that are input and output by the read-aloud sentence generation unit according to the seventh embodiment.

FIG. 28 is a configuration diagram of a reading sentence generation unit according to the eighth embodiment.

FIG. 29 shows a final format conversion document input to the reading sentence generating unit according to the eighth embodiment.

FIG. 30 shows a read-aloud sentence output from a read-aloud sentence generator according to the eighth embodiment.

FIG. 31 is a flowchart of a skipping process of a conventional speech synthesizer.

[Explanation of symbols]

Reference Signs List 10 format conversion unit, 11 heading discrimination unit, 12 paragraph discrimination unit, 13 format conversion document output unit, 14 heading format analysis unit, 20 selection unit, 21 heading selection unit, 22 paragraph selection unit, 23 sentence selection unit, 24 skipping Restricted part, 2
5 sentence boundary keyword detection unit, 26 paragraph addition unit, 30
Text-to-speech generation unit, 31 Text-to-speech output unit, 32
Keyword extraction unit, 33 Modifier clause processing unit, 34 Skipped character number notification unit, 40 voice synthesis unit, 41 control unit,
42 text-to-speech converter, 43 signal sound generator, 44
Adder, 50 input documents, 53 heading format, 60 first format conversion document, 61 second format conversion document, 62
Third format conversion document, 63 Fourth format conversion document, 64
Sentence boundary keyword position information, 70 final format conversion document, 71 keyword, 72 sentence from which adnominal modifier clause is deleted, 73 skipped notice sentence, 80 read-out sentence, 81
Read aloud character string, 82 generation command, 90 output voice, 9
1 Synthetic voice, 92 signal sound.

Claims (12)

[Claims] An input document divided into a plurality of sentences is analyzed, and character information constituting each sentence of a heading item and a paragraph following the heading item in the input document, and the heading item and the paragraph. A format conversion unit that generates and outputs a format conversion document including structural information that can be identified by a user, selects a predetermined sentence in the format conversion document, and adds a final format conversion document obtained by adding this selection information to the format conversion document. A selecting unit that generates and outputs a text-to-speech generating unit that generates and outputs, as a text-to-speech, character information of a sentence to which the selection information is added from the final format conversion document; and a voice that converts the text-to-speech to speech and outputs it. A speech synthesizer, comprising: a synthesizer. 2. The format conversion unit determines a heading item based on the document structure tag from an input document having a document structure tag capable of identifying the heading item and paragraph, and converts the structure information that can identify the heading item from the input document. A heading discriminator for outputting, a paragraph discriminator for discriminating a paragraph from the input document based on the document structure tag, and a paragraph discriminator for outputting structure information capable of identifying the paragraph; and extracting the character information from the input document. 2. The speech synthesis apparatus according to claim 1, further comprising: a format conversion document output unit configured to generate and output a format conversion document including the extracted character information and the structural information that can identify the heading item and the paragraph. 3. The format conversion unit checks a format representing a heading format from the input document, and determines a heading format analysis unit that outputs the format and a character string representing the heading item based on the format. A heading discriminator that outputs structure information that can identify the heading item, a paragraph discriminator that discriminates a paragraph that starts with indentation in a sentence following the heading item, and outputs structural information that can identify this paragraph; 2. The voice according to claim 1, further comprising: a format conversion document output unit configured to generate and output a format conversion document including the character information in the input document and the structural information capable of identifying the heading item and the paragraph. Synthesizer. 4. A final format conversion document in which a heading item up to a predetermined hierarchy is selected from an input document having a plurality of hierarchical heading items and this selection information is added to the format conversion document. 2. The speech synthesis apparatus according to claim 1, wherein the speech synthesis apparatus generates and outputs the speech. 5. A heading selecting unit for selecting a predetermined heading item in the format conversion document and adding this selection information to the format conversion document, and a lower hierarchy of the selected heading item. A paragraph selecting section for selecting a predetermined paragraph to be composed and adding this selection information to the format conversion document; and selecting a predetermined sentence in the selected paragraph and adding this selection information to the format conversion document. The speech synthesis device according to claim 1, further comprising a sentence selection unit. 6. The sentence selection unit according to claim 5, wherein a head sentence and a tail sentence of the paragraph selected by the paragraph selection unit are selected, and the selected information is added to the format conversion document. Voice synthesizer. 7. The sentence selection unit, wherein when the number of characters of a plurality of skipped sentences from a position after the selected sentence to before a next selected sentence exceeds a predetermined value, a plurality of skipped sentences is set. 6. The speech synthesis apparatus according to claim 5, wherein a sentence located in the middle of the text is selected, and the selected information is added to the format conversion document. 8. A sentence boundary keyword detection unit that detects a sentence boundary keyword indicating a break of a series of sentences from a plurality of sentences constituting the paragraph selected by the paragraph selection unit, and outputs the position of the sentence boundary keyword. And a paragraph adding unit that divides the paragraph at the position of the detected sentence boundary keyword and adds a new paragraph, wherein the sentence selecting unit selects the new paragraph added by the paragraph adding unit. 6. The speech synthesizer according to claim 5, wherein said selection information is added to said format conversion document. 9. The reading sentence generating unit includes a keyword extracting unit that checks the appearance frequency of an independent word from the sentence of the paragraph to which the selection information is added, and outputs the independent word having a high appearance frequency as a reading sentence. The speech synthesizer according to claim 1, wherein: 10. The reading sentence generating unit includes a modifying clause processing unit that detects a continuous modifier from a sentence of a paragraph to which the selection information is added, and outputs a sentence from which the continuous modifying clause is deleted as a spoken sentence. The speech synthesizer according to claim 1, wherein: 11. The reading sentence generation unit may be configured such that the number of characters of a sentence to be skipped from after the sentence to which the selection information is added to before to the sentence to which the next selection information is added exceeds a predetermined value. 2. The speech synthesizer according to claim 1, further comprising a skipped character number notifying unit for adding a skipped notification sentence for notifying the number of characters of the skipped sentence to the read-out sentence. 12. The speech synthesizer outputs the character information in the speech sentence output from the speech sentence generator as a speech character string, and further issues an instruction to generate a signal sound in accordance with the boundary of each sentence in the speech sentence. A control unit, a text-to-speech conversion unit that converts the read-out character string output from the control unit to a synthesized voice, a signal sound generation unit that outputs a signal sound according to the generation command, 2. The speech synthesizer according to claim 1, further comprising: an adder for generating and outputting an output speech by adding.