JP2001229180A - Contents retrieval device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently obtain desirable proper contents by automatically extracting a key word by processing input information, such as voice information and image information, through voice recognition and image recognition. SOLUTION: The result of voice recognition by a voice recognition part 2 as to input voice information from a voice input part 1 is outputted from a voice recognition result notification part 3 as a key word and a key word comparison part 4 compares the outputted key word with key words which are stored and managed in a key word-content correspondence table part 5 corresponding to contents to obtain the key word corresponding to the key word as the voice recognition result in the key word-contents correspondence table part 5. The contents corresponding to the detected key word are obtained from the key word-contents correspondence table part 5, and the contents are retrieved from a contents storage part 7 and read out, and then outputted from a contents output part 8.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a content search device for recognizing a keyword included in input information, searching for and outputting content matching the recognized keyword.

[0002]

2. Description of the Related Art Conventionally, in order for a user to search and retrieve desired content, the user himself manually inputs a keyword relating to the content, and retrieves content corresponding to the input keyword.

[0003]

As described above, conventionally, since the user himself manually inputs a keyword and retrieves desired content, it takes time to input a keyword for obtaining appropriate content. However, there is a problem that a great burden is imposed on the user.

[0004] The present invention has been made in view of the above,
The purpose is to extract keywords by voice or image recognition of input information such as voice information and image information,
An object of the present invention is to provide a content search device capable of efficiently obtaining desired appropriate content.

[0005]

According to a first aspect of the present invention, there is provided a voice input unit for inputting voice information, and a voice recognition unit for recognizing voice information input from the voice input unit. Means, a voice recognition result notifying means for outputting a recognition result recognized by the voice recognition means as a keyword, a content storage means for storing a plurality of contents, and a keyword defined for each content corresponding to each content. A keyword content correspondence table means for storing and managing as a table, and comparing a keyword of a speech recognition result output from the speech recognition result notification means with a keyword stored and managed in the keyword content correspondence table means in correspondence with the content. And the keyword corresponding to the keyword in the speech recognition result Keyword comparison means for detecting and outputting from the content correspondence table means, and content corresponding to the keyword output from the keyword comparison means is obtained from the keyword content correspondence table means, and the content is retrieved from the content storage means. The gist of the present invention is to include a control unit for controlling reading and a content output unit for outputting the content read from the content storage unit under the control of the control unit.

According to the first aspect of the present invention, the result of voice recognition of input voice information is used as a keyword, and this keyword is compared with a keyword stored and managed in correspondence with content in the keyword content correspondence table means. Then, a keyword corresponding to the keyword of the voice recognition result is detected from the keyword content correspondence table means, and a content corresponding to the detected keyword is obtained from the keyword content correspondence table means,
Since the content is retrieved from the content storage means, read out and output, the user does not need to manually input the keyword, and an appropriate keyword from which the desired content can be extracted can be automatically extracted by voice recognition. Therefore, the operability of the user can be made more efficient without burdening the user or putting a great burden on the user.

According to a second aspect of the present invention, there is provided an image inputting means for inputting image information, an image recognizing means for recognizing image information input from the image inputting means, and an image recognizing means for recognizing the image information. Image recognition result notifying means for outputting a recognition result as a keyword, content storing means for storing a plurality of contents, and keyword content correspondence table means for storing and managing a keyword defined for each content as a table corresponding to each content And comparing the keyword of the image recognition result output from the image recognition result notifying means with the keyword stored and managed corresponding to the content in the keyword / content correspondence table means, and corresponds to the keyword of the image recognition result. Detect keywords from keyword content correspondence table Keyword comparing means for outputting the contents corresponding to the keyword outputted from the keyword comparing means from the keyword contents correspondence table means, and controlling to retrieve and read the contents from the contents storing means. And a content output unit that outputs the content read from the content storage unit under the control of the control unit.

According to the present invention, the result of image recognition of the input image information is used as a keyword, and this keyword is compared with a keyword stored and managed corresponding to the content in the keyword content correspondence table means. Then, a keyword corresponding to the keyword of the image recognition result is detected from the keyword content correspondence table means, and a content corresponding to the detected keyword is obtained from the keyword content correspondence table means,
Since the content is retrieved from the content storage means, read out, and output, the user does not need to manually input a keyword, and an appropriate keyword from which desired content can be extracted can be automatically extracted by image recognition. Therefore, the operability of the user can be made more efficient without burdening the user or putting a great burden on the user.

Further, according to the present invention, there is provided a voice input means for inputting voice information, a voice recognition means for recognizing voice information input from the voice input means, and a voice recognition means for recognizing the voice information. Voice recognition result notifying means for outputting a recognition result as a keyword, image input means for inputting image information in synchronization with the input of the voice information, and image information input from the image input means for voice information recognition processing Image recognition means for recognizing in synchronization with the image recognition apparatus, image recognition result notifying means for outputting the recognition result recognized by the image recognition means as a keyword, keyword output from the voice recognition result notifying means, and image recognition result notifying means A comparison / decision unit that compares the keywords output from the keyword and determines whether the keywords match, and if the keywords match, the key A keyword selecting means for selecting a keyword of the image recognition result output from the image recognition result notifying means as the target keyword if the word is a target keyword, and a content storing means for storing a plurality of contents; A keyword content correspondence table means for storing and managing a keyword defined in the table as a table corresponding to each content, and a comparison between the target keyword and a keyword stored and managed corresponding to the content in the keyword content correspondence table means A keyword comparison means for detecting and outputting a keyword corresponding to the target keyword from the keyword content correspondence table means; and a content corresponding to the keyword output from the keyword comparison means, Control means for acquiring the content from the content correspondence table means and controlling the content to be retrieved and read from the content storage means; and content output means for outputting the content read from the content storage means under the control of the control means. The gist is to have.

According to the third aspect of the present invention, voice information and image information are input in synchronization, voice recognition of input voice information and image recognition of input image information are performed in synchronization, and a voice recognition result is output. The keyword is compared with the keyword of the image recognition result. If the two keywords do not match, the keyword of the image recognition result is set as a target keyword, and the target keyword and the keyword content correspondence table are stored and managed in correspondence with the content. By comparing the keyword with the keyword, a keyword corresponding to the target keyword is detected from the keyword content correspondence table means, a content corresponding to the detected keyword is obtained from the keyword content correspondence table means, and the content is retrieved from the content storage means. To read and output, the user himself There is no need to input keywords during work, and appropriate keywords that can extract desired content can be automatically extracted by image recognition, improving efficiency of user operability without burdening the user and placing a heavy burden on the user. can do.

According to a fourth aspect of the present invention, in the second or third aspect of the present invention, the image recognizing means includes a telop recognizing means for recognizing a telop from input image information as image information, and The gist of the present invention is to include a cutout unit that cuts out character information and outputs the cutout character information as a recognition result.

According to the present invention, a telop is recognized as image information from input image information, character information is cut out from the recognized telop, and the cut out character information is used as a keyword of an image recognition result. The keyword is compared with the keyword in the keyword content correspondence table means to detect the keyword, the content corresponding to the detected keyword is obtained from the keyword content correspondence table means, and the content is retrieved and read from the content storage means. Output
There is no need for the user to manually input keywords, and appropriate keywords that can extract desired content can be automatically extracted by image recognition. Efficiency can be improved.

The present invention described in claim 5 is the same as the claim 2.
In the present invention according to the third or third aspect, the image recognition means is:
The gist of the invention is to include a caption recognition unit that recognizes a caption as image information from input image information, and a cutout unit that cuts out character information from the recognized caption and outputs the cutout character information as a recognition result.

According to the fifth aspect of the present invention, a caption is recognized as image information from input image information, character information is cut out from the recognized caption, and the cut out character information is used as a keyword of the image recognition result. The keyword is compared with the keyword in the keyword content correspondence table means to detect the keyword, the content corresponding to the detected keyword is obtained from the keyword content correspondence table means, and the content is retrieved and read from the content storage means. Output without the need for the user to manually input keywords, and appropriate keywords that can extract the desired content can be automatically extracted by image recognition, which is troublesome and burdens the user. And improve user operability It is possible.

Further, the present invention described in claim 6 is the first invention.
In the present invention, the content storage means is provided outside the content search device, and the control means accesses the content storage means provided outside the content search device, and The main point is to have an external access means for reading content from.

According to the present invention, the content storage means is provided outside, and the content is read by accessing the content storage means provided outside.

The present invention according to claim 7 provides the invention according to claims 1 to 5
In the present invention described in any one of the above, the gist is that the content storage means is constituted by a removable recording medium.

According to the present invention, the content storage means is constituted by a removable recording medium.

Further, the present invention described in claim 8 provides the present invention according to claim 1.
The present invention according to any one of the first to seventh aspects, wherein the keyword content correspondence table means is constituted by a removable recording medium.

According to the present invention, the keyword content correspondence table means is constituted by a removable recording medium.

Further, the present invention described in claim 9 is based on claim 1.
In the present invention according to the third or third aspect, the recognition result recognized by the voice recognition means is subjected to morphological analysis to determine a local topic range, and local topic words are extracted. Misrecognition determining means for determining whether or not the recognition result is erroneous recognition based on the lower concept relation and the synonym relation; and, as a result of the determination, a synonym of a word having a difference of one character from the recognition result determined to be erroneous recognition. The gist of the present invention is to include a correction candidate output unit that selects a word having a relationship as a target word, and outputs a correction candidate for the recognition result based on the concept distance between the target word and the local topic word.

According to the ninth aspect of the present invention, a speech recognition result is subjected to morphological analysis to determine a local topic range, a local topic word is extracted, and a superordinate / subordinate conceptual relationship with respect to the local topic word is determined. And determining whether or not the recognition result is erroneous recognition based on the synonymous relationship, and selecting a word having a synonymous relationship of a word having a difference of one character from the recognition result determined as erroneous recognition as a target word. To output correction candidates for the recognition result based on the conceptual distance between
Even if the recognition result is erroneous, it is possible to obtain a correct correction candidate keyword and obtain appropriate content.

[0023]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of a content search device that performs a content search method according to the first embodiment of the present invention. The content search device shown in FIG. 1 includes a voice input unit 1 to which voice information is input, a voice recognition unit 2 for recognizing voice information input from the voice input unit 1, and a recognition result recognized by the voice recognition unit 2. Recognition result notification unit 3 that outputs a keyword as a keyword, a content storage unit 7 that stores a plurality of contents, and a keyword content correspondence table that manages a keyword defined for each content as a table corresponding to each content The unit 5 compares the keyword of the speech recognition result output from the speech recognition result notifying unit 3 with the keyword stored and managed corresponding to the content in the keyword content correspondence table unit 5, and corresponds to the keyword of the speech recognition result. Keyword ratio for detecting and outputting keywords from the keyword content correspondence table unit 5 Part 4, the control unit 6 to the keyword content corresponding to the keyword output from the comparison unit 4 acquires from the keyword contents correspondence table unit 5, and controls to read and search for this content from the content storage section 7,
A content output unit 8 outputs the content read from the content storage unit 7 under the control of the control unit 6.

Next, the operation of the first embodiment shown in FIG. 1 will be described with reference to the flowchart shown in FIG.

First, when audio information input from an audiovisual device such as a television or a radio or audio information input via a communication interface or the like is input via the audio input unit 1 (step S100), The input voice information is voice-recognized by the voice recognition unit 2 (step S101), and the voice recognition result is output from the voice recognition result notification unit 3 as a keyword (step S101).
102).

The keyword output from the voice recognition result notifying unit 3 is input to the keyword comparing unit 4 and
In addition, the keyword stored in the keyword content correspondence table unit 5 corresponding to the content is also input to the keyword comparison unit 4 (step S103), and both keywords are compared by the keyword comparison unit 4 and correspond to the keyword of the speech recognition result. The keyword to be executed is detected and output from the keyword content correspondence table section 5 (step S104).

The keyword content correspondence table section 5 is searched for the keyword output from the keyword comparison section 4, and the content corresponding to the keyword is output from the keyword content correspondence table section 5 (step S105). Upon receiving the content output from the keyword content correspondence table unit 5, the control unit 6 accesses the content storage unit 7 (step S10).
6), the content is retrieved and read from the content storage unit 7 (step S107), and the read content is output from the content output unit 8 (step S10).
8).

FIG. 3 is a block diagram showing a configuration of a content search device according to the second embodiment of the present invention. The content search device shown in FIG. 1 is directed to image information while the first embodiment shown in FIG. 1 is directed to audio information. An image input unit 9 for inputting image information by the recognition unit 2 and the voice recognition result notification unit 3, an image recognition unit 10 for recognizing image information input from the image input unit 9, and recognition by the image recognition unit 10 The only difference is that an image recognition result notifying unit 11 that outputs the obtained recognition result as a keyword is different. The other keyword comparing unit 4, keyword content correspondence table unit 5, control unit 6, content storage unit 7, and The content output unit 8 is the same as that of the embodiment of FIG.

The operation of the second embodiment shown in FIG. 3 will be described with reference to the flowchart shown in FIG.

When image information from an audiovisual device or the like is input via the image input unit 9 (step S).
200), the image information is image-recognized by the image recognition unit 10 (step S201), and the image recognition result is output from the image recognition result notification unit 11 as a keyword (step S202).

The keyword output from the image recognition result notifying section 11 is input to the keyword comparing section 4, and the keyword stored in the keyword comparing section 4 in correspondence with the content in the keyword content correspondence table section 5. Are also input (step S203), the two keywords are compared by the keyword comparison unit 4, and a keyword corresponding to the keyword of the image recognition result is detected and output from the keyword content correspondence table unit 5 (step S204).

The keyword content correspondence table section 5 is searched for the keyword output from the keyword comparison section 4, and the content corresponding to the keyword is output from the keyword content correspondence table section 5 (step S205). Upon receiving the content output from the keyword content correspondence table section 5, the control section 6 accesses the content storage section 7 (step S20).
6), the content is retrieved and read from the content storage unit 7 (step S207), and the read content is output from the content output unit 8 (step S20).
8).

FIG. 5 is a flowchart showing a processing procedure of the content search method according to the third embodiment of the present invention. The content search method shown in the figure comprises a content search device combining the functions of both the content search device for audio information shown in FIG. 1 and the content search device for image information shown in FIG. Is what is done.

In the content search processing shown in FIG. 5, first, when audio information and image information from the audiovisual device are input via the audio input unit 1 and the image input unit 9 (steps S300 and S303), respectively. The voice information and the image information are subjected to voice recognition and image recognition in synchronization with the voice recognition unit 2 and the image recognition unit 10, respectively (steps S301 and S304), and the voice and image recognition results are used as keywords as voice recognition result notification units, respectively. 3 and output from the image recognition result notifying section 11 (step S
302,305).

The two keywords output from the voice recognition result notifying unit 3 and the image recognition result notifying unit 11 are compared with each other.
It is determined whether the two recognition results match (step S).
306). If the two keywords match, the keyword is output as the target keyword. If they do not match, the keyword of the image recognition result output from the image recognition result notification unit 11 is output as the target keyword. (Step S308), the keyword comparison unit 4
And the keyword stored in the keyword content correspondence table unit 5 corresponding to the content is also input to the keyword comparison unit 4 (step S309), and both keywords are compared by the keyword comparison unit 4 and recognized. A keyword corresponding to the target keyword is detected and output from the keyword content correspondence table unit 5 (step S310).

The keyword content correspondence table unit 5 is searched for the keyword output from the keyword comparison unit 4, and the content corresponding to the keyword is output from the keyword content correspondence table unit 5 (step S311). Upon receiving the content output from the keyword content correspondence table unit 5, the control unit 6 accesses the content storage unit 7 (step S31).
2) The content is retrieved and read from the content storage unit 7 (step S313), and the read content is output from the content output unit 8 (step S31).
Four).

FIG. 6 is a block diagram showing the configuration of a content search device according to the fourth embodiment of the present invention. The content search device shown in the figure recognizes a telop as image information from input image information in the second embodiment shown in FIG. 3, cuts out character information from this telop, and uses the cut out character information as a keyword of the recognition result. 3 in that a telop recognition unit 12 and a telop recognition result notification unit 13 are provided instead of the image recognition unit 10 and the image recognition result notification unit 11 in the embodiment of FIG. The effect is the same.

That is, the telop recognition unit 1 in FIG.
Reference numeral 2 denotes a telop recognizing unit for recognizing a telop as image information from the input image information from the image input unit 9, and a extracting unit for extracting character information from the recognized telop and outputting the extracted character information as a recognition result. Then, the telop recognition result notifying section 13 outputs the recognition result output from the cutout means as a keyword.

The operation of the fourth embodiment shown in FIG. 6 will be described with reference to the flowchart shown in FIG.

When image information from an audiovisual device or the like is input via the image input unit 9 (step S
400), the telop included in the image information is recognized by the telop recognition unit 12 (step S401), character information is cut out from the recognized telop, and the telop recognition result notification unit 13 outputs the telop recognition result as a keyword. (Step S402).

The keyword output from the image recognition result notifying section 11 is input to the keyword comparing section 4, and the keyword stored in the keyword comparing section 4 in correspondence with the content in the keyword content correspondence table section 5. Are also input (step S403), the two keywords are compared by the keyword comparison unit 4, and a keyword corresponding to the keyword of the telop recognition result is detected and output from the keyword content correspondence table unit 5 (step S404).

The keyword content correspondence table unit 5 is searched for the keyword output from the keyword comparison unit 4, and the content corresponding to the keyword is output from the keyword content correspondence table unit 5 (step S405). Upon receiving the content output from the keyword content correspondence table unit 5, the control unit 6 accesses the content storage unit 7 (step S40).
6), the content is retrieved and read from the content storage unit 7 (step S407), and the read content is output from the content output unit 8 (step S40).
8).

FIG. 8 is a block diagram showing a configuration of a content search device according to the fifth embodiment of the present invention. The content search device shown in the figure recognizes a caption as image information from input image information in place of a telop in the fourth embodiment shown in FIG. 6, cuts out character information from this caption, and converts the cut out character information. The telop recognition unit 12 and the telop recognition result notification unit 1 of the embodiment shown in FIG.
3 in that a caption recognition unit 14 and a caption recognition result notification unit 15 are provided in place of 3.
Other configurations and operations are the same.

That is, the caption recognizing unit 14 in FIG. 8 includes a caption recognizing means for recognizing a caption as image information from the image information input from the image input unit 9, cuts out character information from the recognized caption, and outputs the cut out character. The caption recognition result notifying unit 15 is configured by a cutout unit that outputs information as a recognition result, and outputs the recognition result output from the cutout unit as a keyword.

The operation of the fifth embodiment shown in FIG. 8 will be described with reference to the flowchart shown in FIG.

When image information from an audiovisual device or the like is input through the image input unit 9 (step S
500), the caption included in the image information is recognized by the caption recognition unit 14 (step S501), character information is cut out from the recognized caption, and the caption recognition result is output from the caption recognition result notification unit 15 as a keyword. (Step S502).

The keyword output from the caption recognition result notifying section 15 is input to the keyword comparing section 4, and the keyword stored in the keyword comparing section 4 corresponding to the content in the keyword content correspondence table section 5. Is also input (step S50).
3), both keywords are compared by the keyword comparison unit 4,
A keyword corresponding to the keyword of the caption recognition result is detected and output from the keyword content correspondence table unit 5 (step S504).

The keyword content correspondence table unit 5 is searched for the keyword output from the keyword comparison unit 4, and the content corresponding to the keyword is output from the keyword content correspondence table unit 5 (step S505). Upon receiving the content output from the keyword content correspondence table section 5, the control section 6 accesses the content storage section 7 (step S50).
6), the content is retrieved and read from the content storage unit 7 (step S507), and the read content is output from the content output unit 8 (step S50).
8).

In each of the above-described embodiments, when the recognition result of voice recognition or image recognition is output as a keyword, if the estimated probability is equal to or less than a predetermined threshold, the keyword is not output. Is simply notified to the user. When the number of recognized keywords is a fixed number as described above, the recognition process is repeated by changing the set value, that is, by increasing the reference appearance probability. Note that the estimated probability is called likelihood in speech recognition, for example.
Next, a description will be given of a process of obtaining an estimated probability from voice, video, characters, and the like to be recognized.

In this process, first, as described in each of the above-described embodiments, a feature amount is extracted from input contents such as audio information and image information input from the input unit. For example, if the input information is speech, the feature amount is extracted using LPC (Linear Prediction Analysis) Delta Kept Slam or the like,
In the case of an image such as a telop, the feature amount is a pattern representing a character.

When the feature is extracted, a plurality of symbol candidates corresponding to the feature are determined. This is a symbol (for example, a character or a chain of characters) that features a feature
Is performed by determining the candidate and the appearance probability p _i of the symbol based on the feature amount corresponding relationship between the candidate and the appearance probability p _i from the database that stores a number. Here, i is a code for distinguishing a character string.

Next, the plurality of symbols determined in this way are connected to determine a plurality of keyword candidates. This is performed by, for example, connecting a plurality of determined symbol candidates in the input order to determine a plurality of character strings, that is, a plurality of keyword-like candidates. At this time, the estimated probability of occurrence of each character string candidate is calculated as the product TTi · pi probability of occurrence p _i. Then, words existing by the word dictionary are extracted from the determined character strings. However, the character string and the word usually do not match, and a word of plus or minus several characters is further selected from the determined character string.

Next, from among the plurality of keywords determined as described above, candidate words whose estimated probability is equal to or more than a predetermined threshold are selected as keywords, and the selected keywords are used in each of the above-described embodiments. I use it.

FIGS. 10 and 11 are flowcharts showing the processing procedure of the content search method according to the sixth embodiment of the present invention. The content search method shown in both figures determines and corrects an erroneous recognition part from the recognition result recognized from the input voice information or the input image information in each of the embodiments shown in FIGS. It outputs the recognition result candidates.

That is, in FIGS. 10 and 11, first, as described above, the voice recognition unit 2 and the image recognition unit 1
0, telop recognition unit 12, or caption recognition unit 1
4, the sentence of the recognition result recognized is received, the sentence is subjected to morphological analysis, the sentence is divided into words, and the parts of speech of the words are determined (step S601). Then, the result of the morphological analysis is used as input to perform a preview process, and a global topic list in which a global topic range and a global topic word are paired is extracted (step S602).

In the flash processing, a global topic range, which is a continuing part of a topic, and a global topic word, which is a word representing the topic, are determined based on a specific phrase used when writing a sentence. The general topic is a topic at a higher level of a sentence, for example, a topic in a “chapter” or the like. The output is a global topic list in which a global topic range and a global topic word are paired.

As described above, when a global topic list in which a global topic range and a global topic word are paired is extracted, one global topic list is extracted from the global topic list in which the global topic range and the global topic word are paired. Is selected (step S603). Then, one word is selected from the words included in the selected topic in the general topic range, and this word is set as a target word (step S604). The target word is a word to be determined as to whether the recognition result is correct or incorrect by the following processing, and to correct the error if the result is incorrect.

Next, in order to determine whether the recognition of the target word selected as described above is incorrect or correct,
First, it is checked whether there is a superordinate / subordinate relationship as to whether the general topic is a superordinate concept or a subordinate concept of the concept of the target word (step S605). The determination as to whether the general topic is a superordinate concept or a subordinate concept of the concept of the target word is performed using, for example, a concept tree. The concept tree is obtained by assigning a concept to each node, with the root of the tree structure as the top-level concept. To determine whether there is a superordinate / subordinate relationship between two concepts, check whether there is a node of the other concept on the path from one concept to the root.
If there is a lower order relationship, it is determined that there is no higher order / lower order relationship otherwise. Then, as described above, when the general topic selected from the general topic list is a superordinate concept or a subordinate concept of the concept of the target word, the recognition result is determined to be correct, and the present process ends. If not, the process proceeds to step S606, and it is determined from the synonym dictionary whether or not the general topic and the target word have a synonymous relationship. The synonym dictionary is, for example, as shown in FIG.
One line describes synonymous words. That is, when the general topic and the target word are on the same line of the synonym dictionary, it is understood that there is a synonymous relationship. Then, as a result of the determination in step S606, if the general topic is a synonym of the target word, the recognition result is determined to be correct, and this processing ends.
The process proceeds to the error correction process after step S701.

That is, in the above error determination processing,
If the general topic and the target word of interest have no relation between the superordinate / subordinate concepts and no synonyms, the recognition result is determined to be an error, and the process proceeds to the error correction processing of step S701 and subsequent steps in FIG. Note that such determination is based on the assumption that a topic lower than a certain topic is a topic that is somehow related to a higher topic.

In the error correction process after step S701 in FIG. 11, first, a word having one character difference from the target word of interest is selected from the dictionary (step S701). This is a word that is considered a candidate for a correct recognition result. Then, it is determined whether or not the selected word includes a synonym of the global topic word (step S702). If there is a synonym of the global topic word, the synonym is selected (step S703).

Next, when a synonym is selected as described above, this synonym is included as the currently selected word, and when there is no synonym, the previously selected word is currently selected. A word is extracted from the currently selected word as a target word (step S704). Then, the concept distance between the selected target word and the general topic word is calculated (step S705). The concept distance is the length of the shortest path connecting two concepts on the concept tree. Length is expressed as the number of arcs between concepts.

Then, it is determined whether the concept distance calculated in this way is smaller than the stored minimum concept distance (step S706). If the concept distance is smaller, the concept distance calculated in step S705 is used as the minimum concept distance. It is stored (step S707). Then, the correction candidate of the target word of the current focus is set as the word extracted in step S704 (step S708).

Next, for all the words extracted in step S701, steps S704 to S708 described above are performed.
It is determined whether or not the above processing has been performed, and the same processing is repeatedly performed for all the words (step S709).

Further, it is determined whether or not the above-described processing from step S604 to step S709 has been performed on all target words extracted in step S604, and the same processing is repeatedly performed on all target words (step S71).
0).

Further, it is determined whether or not the processing from step S603 to step S710 has been performed on all the global topics extracted in step S603, and the same processing is repeatedly performed on all the global topics (step S711). As a result of repeating the same process for all global topics, a correction candidate for a misrecognized word is output as the word saved in step S708 (step S712).

A topic is referred to as a preview process from the meaning of the text content, and a topic word is extracted by the preview process in the following method. A topic usually refers to "the subject being talked about about a matter." Here, in particular, a phrase expressing a “subject” corresponding to a matter by a noun phrase in a paragraph is called “topic word”, and a sentence included in the topic word is called “topic sentence”.
Then, the scope of a topic, that is, the area in which a topic is a continuous topic is referred to as the topic scope.
Called. Of course, in a certain "topic scope", the scope of these topics may be nested. Here, the nested structure of the topic scope is referred to as “topic level”. In the present invention, the topic level has two levels. The higher topic level corresponds to a general topic (referred to as a general topic) in which a topic is explicitly presented. On the other hand, a topic that is not explicit, but locally changes the topic (herein referred to as “local topic”) corresponds to a lower topic level.

Next, such topic extraction processing will be described with reference to the flowchart shown in FIG.

In the topic extraction process shown in FIG. 14, a document to be summarized is input as a process target document input process (step S810). In the processing target document input processing, the input document file is opened and the contents of the document are extracted. In step S820, the input document is subjected to morphological analysis of the document to be processed, and morphologically analyzed using the morphological analysis dictionary 31, to determine the word of each sentence described in the document and And morpheme information such as the part of speech and inflected form of each word. The information subjected to the morphological analysis in this way undergoes topic unit determination processing (step S830).

In the morphological analysis, a word dictionary is searched for an input character string, and part-of-speech information (part-of-speech), head-of-sentence availability information (head-of-sentence possible), forward connection information (preceding), backward connection information This is a process of acquiring information such as (subsequent). In a normal word dictionary, a high-speed search can be performed by performing a special dictionary structure called a TREI dictionary structure.
If the dictionary entry includes "Ah", "Greetings", "Ai", etc., the first letter of each (here, because it is Japanese, it is a letter in C language, unlike the alphabet,
Those having the same Japanese characters (indicating 2 bytes of Japanese characters) and those having the same second character are sequentially formed in a tree structure. If the last character matches, the part of speech information (part of speech) and the head
Information such as (beginning of sentence), forward connection information (preceding), and backward connecting information (backward) are described.

The head-of-sentence information is a flag indicating whether or not it can be at the head of the sentence. If the beginning of the sentence is acceptable, the word may be present at the beginning of the sentence, but if the sentence is not the beginning of the sentence, the word is not permitted to be at the beginning of the sentence. The forward connection information is permitted to be connected only when the part of speech or attribute of the preceding word is appropriate, and is deleted as a candidate when the preceding word is not permitted to be connected. Similarly, the backward connection information is permitted to be connected only when the part of speech or attribute of the subsequent word is appropriate, and is deleted as a candidate when the word is not permitted to be connected later. A candidate is selected by such a part-of-speech connection. The maximum likelihood candidate is selected by a method called a minimum cost method.

The minimum cost method is a processing method in which the morpheme candidate having the lowest cost is set as the maximum likelihood candidate. There are two types of costs used in morphological analysis: connection costs and word costs. The connection cost is a cost required for connecting a certain word to another word. Since it is a word and a word, the connection cost for the word + the utilization is zero. The word cost is the cost for that word,
For example, words that are frequently used have lower costs. Also, since utilization is not a word, the cost is zero. By the morphological analysis, the text part is decomposed into words, and at the same time, parts of speech that are considered to be correct are given to each word. In the example of “specify the features of company A” in the topic structure shown in FIG. 13, morphological analysis is performed as shown in Table 1 below.

[0072]

【table 1】 In the next topic unit determination processing (step S830), the sentence described in the document is determined based on the morphological analysis result of the processing target document morphological analysis processing, and the determined sentence is set as a topic unit. In this topic unit determination processing, an input sentence is divided into topic units. The topic unit is a unit that is close to a normal sentence, but is determined to include one normal topic regardless of whether there is a topic. Usually, a sentence is recognized as a topic unit. However, for a sentence in which a topic continuation type clue phrase appears at the beginning of a sentence, a plurality of sentences are regarded as a topic unit, assuming that the topic scope is continued. Thereafter, the process proceeds for each topic unit.

In the next topic marker / topic cue phrase extraction process (step S840), for each topic unit determined in the topic unit determination process, a topic marker which is a follow-up word that explicitly or implicitly indicates the topic Alternatively, a topic cue phrase composed of a connective at the beginning of a sentence, which is a clue for determining a topic, is extracted based on the topic marker / topic cue phrase stored in the topic marker / topic cue phrase storage unit 33.

The topic marker has a postposition role indicating the possibility that a noun phrase that may represent a topic appears. For example, rhetorical phrases that clearly indicate the subject include case particles “ga” and “wo”, particles “ha”, and “about”. The introduction of a topic is determined from clue phrases such as "first" and "first". The sentence logical structure such as “1.” or “Chapter 1” clarifies the scope of a topic. "This,""thisresult," etc., indicate that the topic continues. A question expression such as "I ask about", "or an example expression such as" for example "can be said to be a rhetorical expression for introducing a local topic.

A rhetorical phrase used in determining a topic word is herein called a “topic marker”. Topic markers include:
It is a marker that plays the role of extracting the topic target, and there are explicit markers and implicit markers. Note that the explicit markers of this topic marker include, for example, “about” and “about”, and the non-specific markers include, for example, “ga”, “wo”, and “ni”. However, an explicit marker is a marker that only has the function of presenting a phrase, and an implicit marker is a word that indicates a grammatical role, such as subject or object, such as "ga" or "wo". It is also used for

Although the topic word is not directly specified, there is a topic cue phrase indicating that a sentence or a single sentence may include the topic word. Topic clue phrases are used as clues to detect a semantic unity of a topic.The topic clue phrases include a nested start type, a topic conversion type, a nested end type, and a nested start type. For example, there are "first,""first,""first," and the like.
And "second", and the nested type includes, for example, "end" and "end".

Further, the topic cue phrases include a topic continuation type cue phrase indicating the continuation of a topic, a topic start, end, and conversion type cue phrases. Further, examples of clues for extracting a local topic include: There is a question expression. Examples of this exemplification include “for example”, “that is,” “specifically,” and the like, and are expressions that lead to the specific example in response to the preceding statement. In addition, examples of question expressions are "~?"
This is a phrase that indicates a question, such as "~ masuka". The use of these phrases is specified only after the text is divided into words in advance by morphological analysis, and a part of speech is given to each word.

In the prominent noun phrase extraction process in the next step S850, the prominent noun phrase considered as a topic candidate based on the topic marker and topic cue phrase for each topic unit extracted in the topic marker / topic cue phrase extraction process is described. Extract noun phrases. Prominent noun phrases that are candidates for topic words are selected from words in front of the topic markers. Salient noun phrases are noun phrases that may be identified as topic word candidates,
This is a noun phrase that excludes noun phrases that have no meaning alone, such as pronoun-expressed nouns. Of course, topic clue phrases (topic start / conversion / end / continuation), example expressions, and question expressions themselves are not prominent noun phrases.

Next general topic extraction processing (step S860)
Then, a global topic is extracted based on the topic marker, topic cue phrase, and salient noun phrase extracted in the topic marker / topic cue phrase extraction process and the salient noun phrase extraction process. Topics are extracted using information on these topic markers, salient noun phrases, and topic clue phrases. For example, in the example of the topic structure shown in FIG. 13, the topic marker is indicated by outline characters, the topic clue is indicated by an underline, and the finally extracted topic is indicated by a square. In the first sentence,
A topic marker “in” is extracted. In the second sentence, "first" and "first" are extracted as topic clue phrases.

The general topic is extracted by a procedure of detecting a general topic introduction part and extracting a topic word of the general topic.

In the detection of the global topic introduction part, when there is a topic unit including the first sentence of a sentence or a topic clue phrase, the topic unit is set as a candidate of the general topic introduction part. If a prominent noun phrase is included in the head simple sentence of a topic unit, the topic unit is determined to be a general topic introduction unit. If the salient noun phrase is not in the topic unit, the next topic unit is set as a candidate for the topic introduction unit. The same processing is continued until the end of the sentence, and the general topic introduction part is determined. This is based on the idea that there is always a global topic introduction section around the topic clue phrase. In other words, it is based on the idea that when writing a sentence, if an introductory expression is made with a topic clue phrase, the topic should always be near it.

Next, in the topic word detection of the general topic,
Among the salient noun phrases extracted in the global topic introduction section, a noun phrase that is most appropriate as a topic word is extracted as a topic word. Based on the following priorities, the noun phrase that appears first is the topic word of the general topic.

(1) In the case where a salient noun phrase is presented by an explicit marker, or in the case of a salient noun phrase including a proper noun (2) In the case of a salient noun phrase which does not contain a proper noun If the topic word extracted in the section overlaps, it is not extracted as a topic word.

In the example of the topic structure shown in FIG. 13, "the characteristics of company A" are extracted from the first line and "X industry" is extracted from the second line as the general topic.

In the local topic extraction processing in the next step S870, the topic marker, topic cue phrase and salient noun phrase extracted in the topic marker / topic cue phrase extraction processing and salient noun phrase extraction processing, and further, the general topic Based on the global topics extracted in the extraction process, local topics that occur locally are extracted.

In this local topic extraction, a local topic refers to a case where a topic is locally introduced by an example such as “for example” or a question expression such as “If heard from”. Since the topic occurs locally, a certain number of sentences are required to express the content of the topic. Therefore, in a local topic, the minimum condition is that a topic unit is expressed from more than a certain number of sentences. Among topic units that meet the minimum condition, those with salient noun phrases are considered as local topic candidates, and among the following items with the highest priority, the earliest salient noun phrase that appears first is the topic of the local topic. Words.

(1) When the salient noun phrase is immediately before the question expression, or when the salient noun phrase is immediately after the example expression (2) When the salient noun phrase is presented by an explicit marker, or when the salient noun phrase is Is included in the global topic introduction section. (3) When proper nouns not included in any global topic are included in the salient noun phrase. Of course, local topics must not be included in the global topic introduction section. Must not be the same as the topic word extracted in the immediately preceding global topic introduction process. In the example of the topic structure shown in FIG. 13, the part “for example, the tie-up of company B ...” is extracted as a local topic. Regarding the introduction section of these local topics, only the introduction is checked, and it is not known when the local topic has ended. Conversely, in normal sentences, the introduction of a topic has the property of being relatively clear, but the end is often ambiguous,
The scope of the topic cannot be clearly specified. Therefore, until the next local topic or global topic appears, the scope of this local topic is determined.

A meaningful topic word is extracted by such a method. By applying the semantic topic word extraction process to the input sentence example shown in the topic structure shown in FIG. 13 and extracting the semantic topic word, the input sentence example is searched for a topic marker “about”, and this topic marker “ Is extracted as a topic word.

As described above, by correcting an error in the recognition result, the accuracy of matching is increased, the content is easily selected, many advertisements are distributed from the client, and the user can obtain other types of information. Become like

Next, referring to the flowcharts shown in FIGS. 15 to 19, steps S830, S840, S85 in FIG.
0, S860, and topic unit confirmation processing shown in S870, respectively.
The topic marker / topic clue phrase extraction processing, prominent noun phrase extraction processing, global topic extraction processing, and local topic extraction processing will be described.

First, the topic unit determination processing will be described with reference to the flowchart shown in FIG.

In the topic unit determination processing of FIG. 15, first, a parameter T for counting the number of topic units is set to 1 (step S1001), and it is determined whether or not the input sentence has been read to the end of the sentence (step S1001). S1003). If the reading has been completed up to the end of the sentence, the present process ends. If the reading has not been completed up to the end of the sentence, the process proceeds to step S1005, and the processing from step S1005 is repeated until the processing is read up to the end of the sentence.

In step S1005, the results for each unit obtained in the processing target document morphological analysis processing are sequentially input, and
It is determined whether the sentence ends with a “.” Term (step S1007). If it ends at the "." Period, the portion from the input head to "." Is set as a topic unit, and the parameter T is incremented by 1 (step S1009). If it does not end with ".", It is determined whether or not bullets starting with "." Or a number are continuous (step S101).
1). If the bullets are continuous, the section from the head of the bullet that starts with "." Or a number to the end of the sentence is set as a topic unit, and the parameter T is incremented by 1 (step S1013).

If the list is not continuous, it is determined whether or not the ending form of the sentence end expression is the final form as a result of the morphological analysis (step S1015). If it is in the terminal form, as a result of the morpheme analysis, the part from the terminal form in which the sentence end expression is used to the end of the sentence is set as a topic unit (step S1017).

If it is determined in step S1015 that the character is not the final form, it is determined whether there are a plurality of spaces (half-width or full-width) between the characters (step S1019). If there are a plurality of spaces between characters, a portion from a place where a plurality of spaces (half-width or full-width) exists between the characters to the end of the sentence is set as a topic unit (step S1021).

If there is no space between the characters in the determination in step S1019, it is determined that the sentence is still continued (step S1023), and the process returns to step S1003.
The processing after step S1005 is repeatedly performed. As described above, the topic unit is determined.

Next, the topic marker / cue phrase extraction processing will be described with reference to the flowchart shown in FIG.

In this processing, first, a parameter T for counting the number of topic unit sentences is set to 1 (step S110).
1) It is determined whether or not this T has reached the maximum value (step S1103). If the T value has reached the maximum value, the present process is terminated. S
The processing after 1105 is repeated until T reaches the maximum value.

In step S1105, the topic marker / topic cue phrase stored in the topic marker / topic cue phrase storage unit 33 shown in FIG. 14 is read. Then, the morpheme in the topic unit sentence T is read, and the parameter M for counting the number of morphemes is set to 1 (step S110).
7). Next, it is determined whether the M-th morpheme in the topic unit sentence matches the topic marker and the topic cue phrase (step S1109). If they match, a topic marker / topic cue phrase that matches for each topic unit is accumulated and stored (step S1111).

If they do not match in the determination in step S1109, the next morpheme is checked and the parameter M
Is incremented by 1 (step S1113). Then, it is determined whether or not the parameter M has been completed up to a predetermined maximum value of the morpheme counting parameter (step S1115). The processing is repeated until the value reaches the value. If the processing has been completed, the parameter T is incremented by 1 in order to process the next topic unit sentence, and the process returns to step S1103 (step S1117). T
Is repeated until the parameter reaches a predetermined maximum value of the topic unit sentence counting parameter. As described above, a topic marker / topic cue phrase is extracted.

Next, the salient noun phrase extraction processing will be described with reference to the flowchart shown in FIG.

In the salient noun phrase extraction processing of FIG. 17, first, a parameter T for counting the number of topic unit sentences is set to 1 (step S1201), and this T is a predetermined maximum value of the topic unit sentence counting parameter. Is determined (step S1203). If the maximum value is reached, this process is terminated. If the maximum value is not reached, the next step S1203 is executed.
The processing after 1205 is repeated until T reaches the maximum value.

In step S1205, the topic marker / topic cue phrase extracted as the topic marker / topic cue phrase is read. Then, it is determined whether or not there is a topic marker in the topic unit sentence (step S1207). If there is no topic marker, the process returns to step S1203. If there is a topic marker, the position of this topic marker is set to P (step S1209).

Then, an independent word located before the position P is extracted and recorded (step S1211). Then, it is determined whether or not the processing has reached the beginning of the sentence (step S12).
13) When the head of the sentence is reached, the independence word or the like recorded so far is recorded as a prominent noun phrase, and the process returns to step S1203 (step S1215).

If the sentence is not the head of the sentence, if the preceding adjunct is a noun phrase connecting a noun phrase such as "no", this noun phrase is recorded, and P is (Step S1217). Then, step S12
Return to 11, and repeat the same process up to the beginning of the sentence.

Next, the general topic extraction processing will be described with reference to the flowchart shown in FIG.

In the general topic extraction processing of FIG. 18, first, a parameter T for counting the number of topic units is set to 1 (step S1301), and whether T has reached a predetermined maximum value of the topic unit counting parameter is determined. Is determined (step S
1303) If the value has reached the maximum value, this process is terminated. If the value has not reached the maximum value, the next step S13
The processing after 05 is repeated until T reaches the maximum value.

In step S1305, a topic marker / topic cue phrase extracted by the topic marker / topic cue phrase extraction processing is read, and a topic unit count is calculated each time a topic unit is extracted by the topic marker / topic cue phrase extraction processing. The parameter T is incremented by +1 and the salient noun phrase extracted in the salient noun phrase extraction processing is read (step S1305). Then, it is determined whether the topic unit counting parameter T = 1, that is, whether it is the beginning of a sentence or has a topic cue phrase, and has a prominent noun phrase (step S1307). If not, the process returns to step S1303 to repeat the same processing. If yes, the topic unit indicated by the topic unit counting parameter T is recorded as the global topic introduction unit (step S1303).
1309).

Next, it is determined whether the explicit marker is a prominent noun phrase to be connected later, whether the prominent noun phrase includes a proper noun, and whether the explicit marker has not been recorded as a general topic (step S1311). . If so, the salient noun phrase is recorded as a global topic, and the process returns to step S1303, and the same process is repeated until T reaches a predetermined maximum value of the topic unit parameter (step S1313). It is further determined whether the explicit marker is a prominent noun phrase that is connected later, whether the prominent noun phrase does not include a proper noun, and whether it has not been recorded as a general topic (step S1315). If so, the salient noun phrase is recorded as a general topic, the process returns to step S1303 (step S1317), and the same process is repeated until T reaches the maximum value. The general topic is extracted by the above processing.

Next, the local topic extraction processing will be described with reference to the flowchart shown in FIG.

In the local topic extraction processing of FIG. 19, first, a parameter T for counting the number of global topics is set to 1 (step S1401), and whether this T has reached a predetermined maximum value of the global topic counting parameter is determined. Is determined (step S
1403) If the value has reached the maximum value, this process ends. If the value has not reached the maximum value, the next step S14
The processing after 05 is repeated until T reaches the maximum value.

In step S1405, the topic marker / topic clue extracted in the topic marker / topic cue phrase extraction processing, the salient noun phrase extracted in the salient noun phrase extraction processing, the global topic extracted in the global topic extraction processing, The global topic introduction unit is read, and the global topic counting parameter T is incremented by +1 each time a global topic is extracted by the global topic extraction process (step S1405).

Then, whether T is greater than 1 (T>
1) That is, it is determined whether the sentence is not the head of the sentence and whether the sentence has a prominent noun phrase (step S1407). If not, the process returns to step S1403. If so, it is determined whether the salient noun phrase is immediately before the question expression or immediately after the example expression and has not been recorded as a general topic. It is determined (step S1409). If so, the salient noun phrase is recorded as a local topic, and the process returns to step S1403. Otherwise, the salient noun phrase is post-connected to the explicit marker or the salient noun phrase is , And has not been recorded as a general topic (step S1413). If so, the salient noun phrase is recorded as a local topic, and the process returns to step S1403; otherwise, the proper noun that is not included in any general topic is further included in the salient noun, and It is determined whether or not the topic has been recorded as a general topic (step S1417). If so, the salient noun phrase is recorded as a local topic, the process returns to step S1403 (step S1419), and the same process is repeated until T reaches a predetermined maximum value of the global topic counting parameter. By the above processing, a local topic is extracted.

In the above embodiment, the content search device having the device configuration shown in the figure has been described.
The present invention is, of course, not limited to such a configuration device, and can also be applied to a content search method for executing the processing procedure of the content search device. Further, the processing procedure of such a content search method is recorded as a program on a recording medium, and this recording medium is incorporated in a computer system, and the program recorded on the recording medium is downloaded or installed in the computer system, and the program By operating the computer system in the above, it can be made to function as a content search device that implements the content search method, and by using such a recording medium, its distribution can be enhanced. .

[0115]

As described above, according to the present invention,
The result of voice recognition of the input voice information is used as a keyword, the keyword is compared with the keyword in the keyword content correspondence table means, a keyword corresponding to the keyword of the voice recognition result is detected, and the content corresponding to this keyword is determined as the keyword content. Since the content is obtained from the correspondence table means, the content is retrieved from the content storage means, read out and output, the user does not need to manually input the keyword, and an appropriate keyword from which the desired content can be extracted is obtained by voice recognition. Extraction can be automatically performed, and the operability of the user can be made more efficient without burdening the user or placing a large burden on the user.

Further, according to the present invention, the result of image recognition of the input image information is used as a keyword, and this keyword is compared with the keyword of the keyword content correspondence table means, and the keyword corresponding to the keyword of the image recognition result is used as the keyword. The content is detected from the content correspondence table means, the content corresponding to the keyword is acquired from the keyword content correspondence table means, and the content is retrieved from the content storage means, read out and output, so that the user himself manually inputs the keyword. There is no need to automatically extract an appropriate keyword that can extract desired content by image recognition, and the efficiency of user operability can be increased without burdening the user or placing a large burden on the user.

Further, according to the present invention, the voice recognition of the input voice information and the image recognition of the input image information are performed in synchronization, and the keyword of the voice recognition result is compared with the keyword of the image recognition result. The keyword of the image recognition result is set as a target keyword, the target keyword is compared with the keyword of the keyword content correspondence table means, a keyword corresponding to the target keyword is detected from the keyword content correspondence table means, and the content corresponding to this keyword is detected. Acquisition, retrieval of this content from the content storage means, and reading and outputting the content, there is no need for the user himself to manually input a keyword, and an appropriate keyword from which desired content can be extracted is automatically extracted by image recognition. Can be troublesome and burdensome for users Without imposing, it is possible to enhance the efficiency of the operation of the user.

According to the present invention, a telop is recognized as image information from input image information, character information is cut out from the recognized telop, and the cut-out character information is used as a keyword of an image recognition result. The keyword is detected by comparing with the keyword in the table means, the content corresponding to the keyword is obtained from the keyword content correspondence table means, and the content is retrieved from the content storage means, read out, and output. There is no need to manually enter keywords, and appropriate keywords that can extract the desired content can be automatically extracted by image recognition, making the user's operability efficient without burdening the user and placing a heavy burden. Can be

Further, according to the present invention, a caption is recognized as image information from input image information, character information is cut out from the caption, and the cut-out character information is used as a keyword of an image recognition result. The keyword is detected by comparing with the keyword in the table means, the content corresponding to the keyword is obtained from the keyword content correspondence table means, and the content is retrieved from the content storage means, read out, and output. There is no need to manually enter keywords, and appropriate keywords that can extract the desired content can be automatically extracted by image recognition, making the user's operability efficient without burdening the user and placing a heavy burden. Can be

Further, according to the present invention, the speech recognition result is subjected to morphological analysis to determine a local topic range, a local topic word is extracted, and a superordinate / subordinate conceptual relationship and a synonymous relationship to the local topic word are determined. It is determined whether or not the recognition result is erroneous recognition based on the recognition result, and a word having a synonymous relationship of a word different from the recognition result determined as erroneous recognition is selected as a target word. Since the correction candidate for the recognition result is output based on the conceptual distance of, the correct correction candidate keyword can be obtained even if the recognition result is erroneous, and appropriate content can be obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a content search device that performs a content search method according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing an operation of the content search device of the first embodiment shown in FIG.

FIG. 3 is a block diagram showing a configuration of a content search device according to a second embodiment of the present invention.

FIG. 4 is a flowchart showing an operation of the content search device of the second embodiment shown in FIG.

FIG. 5 is a flowchart illustrating a processing procedure of a content search method according to a third embodiment of the present invention.

FIG. 6 is a block diagram illustrating a configuration of a content search device according to a fourth embodiment of the present invention.

FIG. 7 is a flowchart showing the operation of the content search device of the fourth embodiment shown in FIG.

FIG. 8 is a block diagram showing a configuration of a content search device according to a fifth embodiment of the present invention.

FIG. 9 is a flowchart showing the operation of the content search device of the fifth embodiment shown in FIG.

FIG. 10 is a flowchart showing a part of a processing procedure of a content search method according to a sixth embodiment of the present invention.

FIG. 11 is a flowchart showing a part of a processing procedure of a content search method according to a sixth embodiment of the present invention, the part following FIG. 10;

FIG. 12 is a diagram illustrating an example of a synonym dictionary.

FIG. 13 is a diagram illustrating an example of a topic structure.

FIG. 14 is a flowchart illustrating a topic extraction process.

FIG. 15 is a flowchart showing a topic unit determination process in step S830 of FIG.

FIG. 16 is a flowchart showing topic marker / topic clue phrase extraction processing in step S840 of FIG. 14;

FIG. 17 is a flowchart showing a prominent noun phrase extraction process in step S850 of FIG. 14;

18 is a flowchart showing a general topic extraction process in step S860 in FIG.

FIG. 19 is a flowchart showing a local topic extraction process in step S870 of FIG.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 voice input unit 2 voice recognition unit 3 voice recognition result notification unit 4 keyword comparison unit 5 keyword content correspondence table unit 6 control unit 7 content storage unit 8 content output unit 10 image recognition unit 11 image recognition result notification unit 12 telop recognition unit 13 Subtitle recognition result notification unit 14 Caption recognition unit 15 Caption recognition result notification unit

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Hiroto Inagaki 2-3-1 Otemachi, Chiyoda-ku, Tokyo F-term in Nippon Telegraph and Telephone Corporation (reference) 5B075 NK31 NK54 PP07 PP10 PP25

Claims (9)

[Claims] 1. Speech input means for inputting speech information, speech recognition means for recognizing speech information input from the speech input means, speech recognition for outputting a recognition result recognized by the speech recognition means as a keyword A result notifying unit, a content storing unit that stores a plurality of contents, a keyword content correspondence table unit that stores and manages a keyword defined for each content as a table corresponding to each content, and a voice recognition result notifying unit. A keyword corresponding to the keyword of the speech recognition result is detected from the keyword content correspondence table means by comparing the output keyword of the speech recognition result with a keyword stored and managed in correspondence with the content in the keyword content correspondence table means. And output keyword comparison And control means for acquiring content corresponding to the keyword output from the keyword comparison means from the keyword content correspondence table means, and controlling the content to be retrieved and read from the content storage means; and And a content output means for outputting the content read from the content storage means under the control of (1). 2. An image input unit for inputting image information, an image recognition unit for recognizing image information input from the image input unit, and an image recognition for outputting a recognition result recognized by the image recognition unit as a keyword. A result notifying unit, a content storing unit that stores a plurality of contents, a keyword content correspondence table unit that stores and manages a keyword defined for each content as a table corresponding to each content, and the image recognition result notifying unit. A keyword corresponding to the keyword of the image recognition result is detected from the keyword content correspondence table means by comparing the keyword of the output image recognition result with a keyword stored and managed in correspondence with the content in the keyword content correspondence table means. And output keyword comparison And control means for acquiring content corresponding to the keyword output from the keyword comparison means from the keyword content correspondence table means, and controlling the content to be retrieved and read from the content storage means; and And a content output means for outputting the content read from the content storage means under the control of (1). 3. Speech input means for inputting speech information, speech recognition means for recognizing speech information inputted from the speech input means, and speech recognition for outputting a recognition result recognized by the speech recognition means as a keyword. A result notifying unit, an image input unit for inputting image information in synchronization with the input of the audio information, and an image recognizing unit for recognizing the image information input from the image input unit in synchronization with the recognition process of the audio information Comparing the keyword output from the voice recognition result notifying means with the keyword output from the image recognition result notifying means, and outputting the recognition result recognized by the image recognizing means as a keyword. Comparing means for judging whether or not both keywords match; and, if both keywords match, the keyword However, if they do not match, a keyword selection unit that selects a keyword of the image recognition result output from the image recognition result notification unit as a target keyword, a content storage unit that stores a plurality of contents, A keyword content correspondence table means for storing and managing a keyword as a table corresponding to each content; comparing the target keyword with a keyword stored and managed corresponding to the content in the keyword content correspondence table means; Keyword comparison means for detecting and outputting a keyword corresponding to the target keyword from the keyword content correspondence table means, and content corresponding to the keyword output from the keyword comparison means in the keyword content correspondence table Control means for acquiring the content from the stage and controlling the content to be retrieved and read from the content storage means; and content output means for outputting the content read from the content storage means under the control of the control means. A content search device characterized by the following. 4. The telop recognition means for recognizing a telop as image information from input image information, and a cutout means for cutting out character information from the recognized telop and outputting the cutout character information as a recognition result. The content search device according to claim 2, further comprising: 5. The image recognizing means includes: a caption recognizing means for recognizing a caption as image information from input image information; and a extracting means for extracting character information from the recognized caption and outputting the extracted character information as a recognition result. 3. The method according to claim 2, wherein
Or the content search device according to 3. 6. The content access means is provided outside the content search device, and the control means accesses the content storage means provided outside and reads out content from the content storage means. The content search device according to any one of claims 1 to 5, further comprising: 7. The content search device according to claim 1, wherein said content storage means is constituted by a removable recording medium. 8. The content search device according to claim 1, wherein said keyword content correspondence table means is constituted by a removable recording medium. 9. A morphological analysis of the recognition results recognized by the voice recognition means and the image recognition means, a local topic range is determined, and local topic words are extracted. An erroneous recognition determining means for determining whether or not the recognition result is erroneous recognition based on a lower concept relation and a synonym relation; 2. A correction candidate output unit that selects a word having a synonym relation as a target word, and outputs a correction candidate for the recognition result based on a conceptual distance between the target word and the local topic word. 4. The content search device according to 2 or 3.