JPH1152973A - Document reading method - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To improve the intelligibility of a document when reading an electronic document, and to enable random access to a document reading position. SOLUTION: The contents of the electronic document are hierarchized by specific information and / or logical structure so that logical characteristics such as titles and paragraphs of the electronic document and / or large visual characteristics of characters can be recognized. The data is converted into internal data, and a reading process is performed based on the internal data. Further, the present invention provides a document reading system that can freely access a reading block of a document in accordance with a user's instruction by using internal data obtained by layering the contents of the document.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to an SGML, HTM
For example, a system for converting an electronic structured document such as L. to a speech by a speech synthesis technique or the like, such as accessing a database by telephone, or having a visually impaired person
Used to access eb).

[0002]

2. Description of the Related Art Conventionally, there are the following methods for reading a text (document) on the premise of visual display. (1) When characterizing text, a character decoration is expressed by a sound, and a reading position in a document is expressed by a sound image localization (referred to as prior art 1, for example, Japanese Patent Laid-Open No. 8-263260)
Publication No. “Text-to-speech method”). (2) When returning the reading position, in the opposite direction for each phrase unit,
In addition, in a phrase unit, a user reads aloud in the forward direction at a high speed, and a user listens to the phrase and searches for a position to be heard back (referred to as a prior art 2; for example, Japanese Patent Laid-Open Publication No. 6-308998, "Speech-to-speech device"). (3) When reading out the document information classified into the hierarchical structure,
A device in which a reading position is designated by a switch pressing time or the number of times (referred to as prior art 3, for example, Japanese Patent Laid-Open No. 5-281)
987, 281988, 281992, “Portable document reading device”). (4) A device which reads out an HTML document with synthesized speech (referred to as a prior art 4, which is commercialized as a software product, for example, "Talking Speech" manufactured by Ricoh Co., Ltd.).

[0003]

However, as described below, these conventional document reading techniques are still insufficient in terms of intelligibility and random access to the read document. This point will be described with reference to the drawings. For example, the HTML document shown in FIG. 1 is visually displayed as shown in FIG. 2 by software generally called a browser. As is clear from FIG. 2, the visibility is improved because the tag information is converted into visual format information to clarify the hierarchical structure in the document. However, if this is converted to speech by the conventional technology 4, the Japanese language is read aloud ignoring the tag in the conventional technology 4 as shown in FIG. 3, so that it becomes difficult for the user to understand. On the other hand, in the above-mentioned prior art 1, since the character modification and the reading position in the document are expressed when converting to speech, visual characteristics can be expressed to some extent by sound, but it is not possible to grasp the logical structure of the document. It is difficult, and grasping the whole picture is in the vague form of sound image localization.

[0004] In the document reading system, there is a problem that random access is difficult due to lack of listability of audio media. In the prior art 3, random access by button operation is realized by adopting a hierarchical file structure. . However, in the prior art 3, existing data cannot be targeted because data is described in a dedicated structure. On the other hand, in the prior art 2, the access speed is obtained by moving through the text at high speed. However, when the document is to be heard back, the reading position of the document is determined while listening to the arrangement of each clause of the document in reverse order. It is hard to say that the load is light for the user.

It is therefore an object of the present invention to improve the intelligibility by preserving the hierarchical structure of a document created mainly for visual use and converting it into speech. A second object of the present invention is to make it applicable to existing structured electronic text as it is in addition to the first object of the present invention. A third object of the present invention is to be applicable to an electronic text without a structured tag in addition to the first object of the present invention. An object of the invention of claim 4 is to further improve the intelligibility of the invention of claim 1. An object of the invention of claim 5 is to make it easy to grasp the position of the listening part in the whole, in addition to the object of the invention of claim 1.

A sixth object of the present invention, in addition to the first object of the present invention, is to quickly perform random access to a read document.

[0007]

SUMMARY OF THE INVENTION The present invention relates to an existing structured electronic text (document) written primarily for visual presentation.
Is converted into a speech including the structure information of the sentence intended by the author, thereby improving the intelligibility and accessibility close to a visual display.

According to a first aspect of the present invention, there is provided a text-to-speech system in which character information of an electronic document is converted into speech by a speech synthesis technique and output, and the contents of the document are hierarchically determined from format information and / or a logical structure of the document. This is a text-to-speech system in which the output order and / or the audio attributes are changed according to the above and converted to speech.

According to a second aspect of the present invention, in the first aspect of the present invention, the format information is treated as a delimiter in a document having a different strength for each type, and thereby the document is hierarchically blocked. It is.

According to a third aspect of the present invention, in the first aspect of the present invention, the first sentence unit of the block, such as the first paragraph in the document and the first sentence in the paragraph, is placed higher than the subsequent sentence units. This is a document reading system in which the contents of the document are hierarchically blocked by positioning them in a hierarchy.

A fourth aspect of the present invention is the document reading system according to the first aspect of the present invention, in which the contents of the document are read in order from a higher hierarchy to a lower hierarchy.

According to a fifth aspect of the present invention, in the first aspect of the present invention, there is provided a text-to-speech system in which a title, a content, and a hierarchy are identified by using audio attributes, additional sounds, and the like.

According to a sixth aspect of the present invention, there is provided a document reading system according to the first aspect, wherein the reading means is movable between upper and lower hierarchies of the document and between the same hierarchies according to a user's request.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described using an HTML document shown in FIG. 1 as an example. FIG. 4 shows this HT
It shows the internal data when the document structure information is extracted from the ML document and converted into hierarchical internal data. FIG. 5 shows a processing flow at the time of the conversion. In the following, first, the conversion process from the source text to the internal data will be described, and then the reading process of the internal data will be described.

The source text is hierarchized mainly based on tags. For this purpose, for example, as shown in FIG. 6, the layers of the tags are ranked in advance from the virtual uppermost layer tag of the uppermost layer to <BR> of the forced line feed at the lowermost layer. This rank is the height of the hierarchy represented by the tag, and is determined by judging the strength as a document break from logical meaning and visual characteristics. The logical meaning is a semantic structure of a sentence mainly represented by a logical tag, such as a title or a paragraph. Visual features are structural information that is visually perceived, such as the size of characters and ruled lines. In the following blocking, the document is divided hierarchically considering these characteristics as the strength of the delimitation of the tag. In this embodiment and the following description, the start tag is mainly used as a block delimiter, and the end tag is not used as a block delimiter. However, the present invention is not limited to this. It is conceivable to treat the range enclosed by tags as a block.

In the following, referring to FIG. 5, the case where the source text shown in FIG. 1 is converted into the hierarchical data shown in FIG. 4 will be described as an example to recursively perform the block division processing of FIG. First, the entire source text of FIG. 1 is set as a processing target text, and a root block having a virtual top layer tag as a current layer tag is divided into blocks (S101 to S101).
111). Title and content extraction processing (S10
2) differs depending on the type of the current hierarchical tag, but is not performed on the virtual uppermost tag. Next, child block CHILD
As a preparation for blocking the text, a tag closest to the current hierarchical level and lower than the current hierarchical level included in the text to be processed is searched (S103 to S103).
S105). Here, the first title tag <H1> is found. Therefore, the child block is blocked for the tag <H1>. That is, a recursive call of the blocking process is performed using the entire source text of FIG. 1 as a processing target text and a tag <H1> as a current hierarchical tag (S101).

As for the tag <H1>, a portion between the tag <H1> and the tag </ H1> is extracted as a title, and the subsequent first paragraph is extracted as the content (S102). As a result, "Speech Web browser TelMePage" is extracted as the title, and the content is not extracted. Next, child block CHIL
The tag of D is searched from the candidate tags lower than <H1>,
Two title tags <H2> are found (S103 to S10)
5). Therefore, the first tag <H2> and the next tag <H
2> as the text to be processed and the tag <H
2> is set as the current hierarchical tag, recursive call of the blocking process is performed (S101 to S107), and this is set as a child block. Further, the same processing is performed from the next tag <H2> to the end (S101 to S107).

In this way, hierarchical block formation is performed recursively (S101 to S111). This processing is repeated except for the part for extracting the title and the content. Therefore, only the title / content extraction processing will be described below. In the tag <H2>, title / content extraction processing is performed in the same manner as the tag <H1>, and the subsequent tag <P> is a child block without a title. This is due to the fact that the first paragraph often gives an outline after that in many cases. The tag <UL> is ignored here, and the tag <L1> portion is processed as a block without a title. Thus, the conversion from the source text to the internal data is performed.

Next, an embodiment of a process for reading out the converted internal data will be described with reference to FIG. When there is no operation from the user, the reading is performed in the order of the title, the content, and the title of the child block (S201 to S210), and the process proceeds to the child block and is similarly read (in the order of the title, the content, and the title of the child block) (S212 to S21).
4). At this time, a different sound effect is added before each of the title, the content, and the title of the child block, and / or the voice type is changed to assist identification.
This is repeated to the lowermost block (S201).
After that, the process returns to the unread block and continues reading out (S202 to S215). As a result, at the beginning of the text-to-speech, the text is composed entirely of the background and the features, and the intelligibility is improved.

Further, during the reading operation, a location designation by an interrupt from the user can be accepted, thereby enabling random access. In other words, by instructing "again" (S301), it is possible to return to the beginning of the block currently being read out and listen again (S2).
03 to S214). For example, when reading out the child block title “feature” of the tag <H1> block,
When "down" is instructed (S302), the reading location can be skipped directly to the contents of "feature" (S212). If "next" is instructed while reading out the contents of the tag <H2>"feature" (S303), it is possible to jump to the child block next to the parent block, that is, the tag <H2> block "background". Similarly, the (U) block in the itemized list can jump to the next item during reading. When "to the upper layer" is instructed (S304), it is possible to return to the head of the parent block and read out.

The above reading operation can be realized by using the processing (S201 to S215) shown in the flowchart of FIG. 7 recursively. In the figure, blockTITLE is the title of the block, blockCONTENT is the content of the block, blockNCHILD is the number of child blocks of the block, and blockCH is used.
ILD [i] represents the i-th child block of the block block. Note that the text-to-speech processing is not described here because existing text-to-speech synthesis technology can be used.

[0022]

According to the first aspect of the present invention, since the information of a document can be mapped and output as a spoken expression while preserving the hierarchical structure, the intelligibility can be improved. Effect corresponding to claim 2: In addition to the effect corresponding to claim 1, the present invention can be applied to existing structured electronic text as it is. Effect corresponding to claim 3: In addition to the effect corresponding to claim 1, the present invention can be applied to an existing electronic text without a structured tag. Effect corresponding to claim 4: In addition to the effect corresponding to claim 1, the intelligibility is improved as a whole by always acquiring in advance the superordinate concept which is the prerequisite knowledge for listening to the subordinate concept. . Effect corresponding to claim 5: In addition to the effect corresponding to claim 1, the position of the place currently being listened to in the entire document is easily grasped, and the intelligibility is improved. Effect corresponding to claim 6: In addition to the effect corresponding to claim 1, it is possible to supplement the difficulty of random access, which is a drawback of audio media, and improve the response speed to user requests.

[Brief description of the drawings]

FIG. 1 illustrates an example of a reading document.

FIG. 2 is a diagram showing a display example of the document by a browser.

FIG. 3 is a diagram showing an example of a read-aloud display when the document is converted into a voice by a conventional method.

FIG. 4 is a diagram showing hierarchical internal data of the document.

FIG. 5 is a diagram showing a processing flow for converting the structure of the document into hierarchical internal data.

FIG. 6 is a diagram showing an example of a tab added to the document.

FIG. 7 is a diagram showing a document reading process flow based on hierarchical internal data.

Claims (6)

[Claims] 1. A document reading system in which character information of an electronic document is converted into speech by a speech synthesis technique and output.
A text-to-speech system characterized by hierarchically grasping the contents of the document from the format information and / or logical structure of the document, and changing the order and / or the voice attribute of the output sentence according to the hierarchically-spoken text. . 2. The document reading system according to claim 1, wherein the format information is treated as a break in a document having a different strength for each type, and the document is hierarchically blocked. . 3. The contents of the document are hierarchically arranged by positioning the first sentence unit of the block, such as the first paragraph in the document and the first sentence in the paragraph, at a higher level than the subsequent sentence units. 2. The method according to claim 1, wherein the document is read out. 4. The document reading system according to claim 1, wherein the contents of the document are read out in order from a higher layer to a lower layer. 5. The document reading system according to claim 1, wherein discrimination between the title, the content, and the hierarchy is performed using a voice attribute, an additional sound, or the like. 6. The document reading system according to claim 1, wherein the reading means is movable between upper and lower layers of the document and between the same layers according to a user's request.