JP2012022053A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically control voice recognition parameters for content.SOLUTION: A voice recognition device includes a content acquisition section 107 which acquires content including voice data and a web page acquisition section 102 which acquires a web page for providing content. The voice recognition device includes a web page analysis section 103 which performs analysis based on the web page for providing content and extracts text indicating characteristics of the voice data. The voice recognition device includes a parameter control section 106 which controls voice recognition parameters for the voice data based on the extracted text and a voice recognition section 111 which performs voice recognition of the voice data in accordance with the controlled voice recognition parameters.

Description

  Embodiments of the present invention relate to speech recognition.

  The user can use various contents via a broadcast wave, a recording medium, or a network (for example, a moving image sharing site). In addition, content reproduction apparatuses are diversified. Specifically, not only a TV receiver but also a mobile phone, a personal computer, a video game machine, and the like may have a content reproduction function.

  It has been proposed to perform speech recognition on audio data included in content and use the speech recognition result as subtitles. Speech recognition is realized using speech recognition parameters such as an acoustic model, a language model, and a word dictionary. In order to obtain a highly accurate speech recognition result, it is important to appropriately control speech recognition parameters for speech data to be recognized. For example, for speech recognition of broadcast programs (mainly news programs), manual speech recognition parameter control (such as learning of an acoustic model and a language model) is performed.

JP 2004-333738 A

  It is inconvenient to manually control speech recognition parameters for each content. On the other hand, if the speech recognition parameters are fixed, it is difficult to perform highly accurate speech recognition for various contents.

  Accordingly, embodiments of the present invention are directed to automatically controlling speech recognition parameters for content.

  A speech recognition apparatus according to an aspect includes a content acquisition unit that acquires content including audio data, and a Web page acquisition unit that acquires a Web page that provides the content. This speech recognition apparatus performs analysis based on a web page that provides content, extracts a text indicating the characteristics of speech data, and converts speech recognition parameters for speech data into the extracted text. And a parameter control unit that performs control based on the control unit. The speech recognition apparatus includes a speech recognition unit that performs speech recognition on speech data according to controlled speech recognition parameters.

  A speech recognition apparatus according to another aspect includes a content acquisition unit that acquires content including audio data. The speech recognition apparatus acquires a web page related to content based on at least one of a speech recognition result of speech data, an image recognition result of video data separated from the content, and text data separated from the content. Includes an acquisition unit. This speech recognition apparatus performs analysis based on a web page related to content, extracts a text indicating the characteristics of the speech data, and converts speech recognition parameters for the speech data into the extracted text. And a parameter control unit that performs control based on the control unit. The speech recognition apparatus includes a speech recognition unit that performs speech recognition on speech data according to controlled speech recognition parameters.

1 is a block diagram showing a voice recognition device according to a first embodiment. The block diagram which shows the speech recognition apparatus which concerns on 2nd Embodiment. The block diagram which shows the speech recognition apparatus which concerns on 3rd Embodiment. The block diagram which shows the speech recognition apparatus which concerns on 4th Embodiment. The block diagram which shows the speech recognition apparatus which concerns on 5th Embodiment. The block diagram which shows the speech recognition apparatus which concerns on 6th Embodiment. Explanatory drawing of an analysis parameter. Explanatory drawing of a control parameter.

Embodiments of the present invention will be described below with reference to the drawings.
(First embodiment)
As shown in FIG. 1, the speech recognition apparatus according to the first embodiment includes a recognition target input unit 101, a web page acquisition unit 102, a web page analysis unit 103, an analysis parameter storage unit 104, an extracted text processing unit 105, a speech. It has a recognition parameter control unit 106, a content acquisition unit 107, a content analysis unit 108, a content separation unit 109, a voice input unit 110, a voice recognition unit 111, and a recognition result output unit 112.

  The recognition target input unit 101 inputs, to the Web page acquisition unit 102 and the content acquisition unit 107, an identifier of a Web page that provides content including audio data that is a target of voice recognition. The identifier of the Web page is expressed in the form of URL (Uniform Resource Locator) or URI (Uniform Resource Identifier), for example.

  The web page acquisition unit 102 acquires a web page according to the web page identifier from the recognition target input unit 101. The web page acquisition unit 102 inputs the acquired web page to the web page analysis unit 103.

  The web page analysis unit 103 performs analysis based on the web page from the web page acquisition unit 102. Specifically, the Web page analysis unit 103 acquires an analysis parameter, which will be described later, from the analysis parameter storage unit 104, and performs analysis according to the analysis parameter. Through this analysis process, the web page analysis unit 103 extracts text indicating the characteristics (acoustic characteristics, linguistic characteristics, etc.) of the speech data that is the target of speech recognition. The web page analysis unit 103 inputs the extracted text to the extracted text processing unit 105.

  In the analysis parameter storage unit 104, for example, analysis parameters are stored in the format shown in FIG. In the example of FIG. 7, the analysis parameter storage unit 104 stores an identifier of a Web page and an analysis parameter in association with each other. In the example of FIG. 7, the analysis parameter is associated with the identifier of a specific Web page, but is common to the identifiers of all Web pages or the identifiers of all Web pages excluding a specific Web page. These analysis parameters may be associated with each other.

  In the example of FIG. 7, the analysis parameter includes a narrowing condition and an application target of the narrowing condition. However, these parameters may not be specified for some or all of the Web page identifiers. The narrowing-down condition is a condition for determining whether or not to perform analysis based on the input Web page. The application target of the narrowing condition is a range that is referred to in order to determine whether or not the narrowing condition is satisfied in the input Web page. For example, if the identifier of the Web page is “http://xxxx.ne.jp”, the Web page analysis unit 103 uses “<title> (. +) </ Title>” parentheses in the source code of this Web page. It is determined whether or not the character string “news” is included in the range indicated by (701). Here, the symbol (. +) Is a regular expression representing one or more occurrences of an arbitrary character (that is, an arbitrary character string) as used in UNIX (registered trademark) or various programming languages. To do. The Web page analysis unit 103 performs analysis based on the Web page if the narrowing-down conditions are satisfied, and omits otherwise. In the example of FIG. 7, another narrowing condition is also associated with the Web page identifier “http://xxxx.ne.jp”. Therefore, the Web page analysis unit 103 includes the character string “Sports” or the character string “Variety” in the range indicated by the parentheses “<genre> (. +) </ Genre)” in the source code of the Web page. It is determined whether or not (702). Further, if the identifier of the Web page is “http://yyyy.ne.jp”, the Web page analysis unit 103 uses “<title> (. +) </ Title>” parentheses in the source code of this Web page. It is determined whether or not the character string “XXX” is included in the range indicated by (703), or the range indicated in parentheses “(. +) Program” in the source code of this Web page It is determined whether or not the character string “□□□□” is included in (704). In the example of FIG. 7, the filtering condition application target is defined by the data structure of the Web page such as the arrangement position of the HTML element or the tag, or by a specific character string. I can't.

  The Web page analysis unit 103 performs analysis based on the input Web page if the narrowing-down conditions are satisfied. Specifically, the Web page analysis unit 103 extracts text from a text analysis target corresponding to the Web page identifier according to an extraction method corresponding to the Web page identifier. The text analysis target is an input web page source code (701, 703), another web page source code (702, 704), and the like. The technical significance of analyzing the source code of another Web page instead of the input Web page is that the input Web page cites content provided in another Web page, etc. It is possible to expect more detailed information. There are various text extraction methods such as full text extraction (701), extraction of a part including a specific character string (703), extraction of a specific part defined by the data structure of the Web page (702, 704). The web page analysis unit 103 may extract a plurality of texts without being limited to one.

  The extracted text processing unit 105 converts the extracted text from the web page analysis unit 103 into control parameters. The control parameter is used for controlling the speech recognition parameter by the speech recognition parameter control unit 106 described later. The speech recognition parameter includes, for example, an acoustic model, a word dictionary, or a language model. The acoustic model represents acoustic features such as phoneme or syllable frequency patterns. The word dictionary lists recognizable word information (notation information, part of speech information, etc.). The language model represents linguistic features such as connection relationships between words. The voice recognition unit 111 described later performs voice recognition based on the acoustic features of the voice data and the linguistic features of the recognition candidates, and generates a recognition result.

  Specifically, the extracted text processing unit 105 performs processing according to the extracted text and the extracted text processing method specified by the Web page analysis unit 103. Typically, the extracted text processing unit 105 performs morphological analysis on the extracted text. The extracted text processing unit 105 converts these morphological analysis results into control parameters and inputs them to the speech recognition parameter control unit 106. For example, the extracted text processing unit 105 may detect a predetermined keyword from the morphological analysis result and directly convert it as a control parameter such as a speaker, a genre, a language, a dialect, or a situation. It may be converted into control parameters using a service, or may be converted into control parameters using an ontology dictionary. Further, if the extracted text processing unit 105 can acquire the temporal correspondence between the content reproduction time and the control parameter, the extracted text processing unit 105 may specify a temporal range to which the control parameter is applied. For example, a comment or the like may be described in association with the playback time of the content on a Web page that provides the content. The extracted text processing unit 105 may convert some or all of the keywords into dictionary parameters. The speech recognition parameter control unit 106 registers word information (notation information, part-of-speech information, etc.) corresponding to the dictionary parameter in the word dictionary, or sets a high recognition priority for this word. Further, the extracted text processing unit 105 may convert the extracted text into language model parameters. The language model parameters can be used for updating (learning) the language model. Note that the dictionary parameter and the language model parameter are part of the control parameter.

  The control parameters are illustrated in FIG. The control parameters include various attributes such as speaker, genre, language, dialect or situation, for example. “Control parameter: speaker” may indicate the generation and / or gender of a speaker such as “male in 20s” and “female in 20s”, or a specific individual (announcer, actor, An actress). “Control parameter: speaker” is effective in selecting an acoustic model, a language model, a word dictionary, and the like. For example, if “control parameter: speaker” is “female”, the speech recognition parameter control unit 106 can select an acoustic model or the like for women. If “control parameter: speaker” indicates a specific individual, the speech recognition parameter control unit 106 may select an acoustic model, a language model, a word dictionary, or the like optimized for the specific individual. it can.

  “Control parameter: genre” indicates a genre of content such as “news”, “drama”, “variety”, and the like. “Control parameter: genre” is effective for selecting an acoustic model, a language model, a word dictionary, and the like. For example, if “control parameter: genre” is “news”, the speech recognition parameter control unit 106 can select an acoustic model, a language model, and a word dictionary for news.

  “Control parameter: language” indicates a language used by a speaker such as “Japanese”, “English”, “Chinese”, and the like. “Control parameter: language” is effective for selecting an acoustic model, a language model, a word dictionary, and the like. “Control parameter: dialect” indicates a dialect corresponding to a subset of the above-mentioned “control parameter: language” such as “standard language”, “Kansai dialect”, “Kyushu dialect”. Therefore, normally, if “control parameter: dialect” is found, “control parameter: language” is also found. Therefore, if “control parameter: dialect” is found, the extracted text processing unit 105 sets the corresponding control “parameter: language”. It may be determined automatically.

  “Control parameter: situation” indicates an audio data recording environment such as “inside a train”, “quiet location”, “inside a car”, and the like. “Control parameter: situation” is effective for selecting an acoustic model, controlling noise canceling processing, and the like.

  As illustrated in FIG. 7, the text analysis target, the text extraction method, and the extracted text processing method are not limited to one set for a specific Web page identifier, and a plurality of sets may be provided (701). For example, the extracted text processing unit 105 may return the top keyword to the web page analysis unit 103, and the web page analysis unit 103 may generate a search expression including the top keyword. The Web page analysis unit 103 may transmit this search formula to a predetermined search engine and extract text based on one or more searched Web pages. When extracting text based on a plurality of searched Web pages, individual priorities may be assigned to the extracted text. The priority may be determined by the identifier of each Web page, or may be determined by the sort order of each Web page in the search result. The priority can be used to select a valid control parameter when the number of control parameters is excessive. In addition, as illustrated in FIG. 7, the text extraction method may be further switched according to the identifier of the searched Web page. In addition, a web page searched based on the head keyword or the like may be handled as a web page newly input to the web page analysis unit 103.

  The speech recognition parameter control unit 106 controls speech recognition parameters for speech data to be recognized according to the control parameter from the extracted text processing unit 105. For example, the speech recognition parameter control unit 106 roughly selects an acoustic model, a language model, and a word dictionary according to “control parameter: language” or “control parameter: dialect”, and “control parameter: speaker” or “control parameter: genre”. The acoustic model, language model, and word dictionary can be selected in more detail. Further, the voice recognition parameter control unit 106 may select an acoustic model in more detail according to “control parameter: situation”, or may control noise canceling processing. The speech recognition parameter control unit 106 may register the word indicated by the dictionary parameter in the word dictionary, or set the recognition priority of this word high. The speech recognition parameter control unit 106 may update the language model according to the language model parameter. The voice recognition parameter control unit 106 may update the language model or the word dictionary temporarily or continuously. That is, the speech recognition parameter control unit 106 may invalidate or validate the update of the language model or the word dictionary after the corresponding speech recognition process is finished. Whether the update of the language model or the word dictionary is temporary or continuous may be determined in advance, may be specified by a control parameter, or individually determined for each language model or each word dictionary Also good.

  The content acquisition unit 107 acquires content corresponding to the identifier of the Web page from the recognition target input unit 101. The content acquisition unit 107 inputs the acquired content to the content analysis unit 108.

  The content analysis unit 108 analyzes the content from the content acquisition unit 107. The content analysis unit 108 extracts metadata and media data from the content, and inputs the content to the content separation unit 109.

  The content separation unit 109 separates audio data included in the media data from the content analysis unit 108. The content separation unit 109 inputs the separated audio data to the audio input unit 110.

  The voice input unit 110 converts the voice data from the content separation unit 109 into a format suitable for the voice recognition unit 111. The voice input unit 110 inputs the converted voice data to the voice recognition unit 111.

  The voice recognition unit 111 performs voice recognition on the voice data from the voice input unit 110 according to the controlled voice recognition parameter after the processing of the voice recognition parameter control unit 106 is completed for the voice data to be recognized. Do. The voice recognition unit 111 inputs the recognition result to the recognition result output unit 112.

  The recognition result output unit 112 outputs a recognition result. For example, the recognition result output unit 112 may display the recognition result as a caption on a display unit (not shown) in synchronization with the reproduction of the content, or save the recognition result as content metadata on a storage medium (not shown). Alternatively, the recognition result may be used for scene detection of content.

  As described above, the speech recognition apparatus according to the first embodiment controls speech recognition parameters based on the Web page that provides content. Therefore, the speech recognition apparatus according to the present embodiment can automatically control speech recognition parameters for content.

(Second Embodiment)
As shown in FIG. 2, the speech recognition apparatus according to the second embodiment includes a speech recognition parameter control unit 106 as a speech recognition parameter control unit 206 and a content separation unit 109 as a content separation unit 209 in the speech recognition apparatus of FIG. This corresponds to a configuration in which a video input unit 213 and an image recognition unit 214 are added. In the following description, the same parts in FIG. 2 as those in FIG. 1 are denoted by the same reference numerals, and different parts will be mainly described.

  The content separation unit 209 separates audio data and video data included in the media data from the content analysis unit 108. The content separation unit 209 inputs the separated audio data to the audio input unit 110. The content separation unit 209 inputs the separated video data to the video input unit 213.

  The video input unit 213 converts the video data from the content separation unit 209 into a format suitable for the image recognition unit 214. The video input unit 213 inputs the converted video data to the image recognition unit 214. Note that the video input unit 213 may thin out frames in the video data from the content separation unit 209 in order to omit image recognition for some frames in the video data.

  The image recognition unit 214 performs image recognition on the video data from the video input unit 213. The image recognition unit 214 generates the above-described control parameter based on the recognition result, and inputs it to the voice recognition parameter control unit 206. Specifically, the image recognition unit 214 may recognize text (for example, a telop or the name of a program performer) displayed in the video and perform morphological analysis on the text. The image recognition unit 214 converts these morphological analysis results into control parameters and inputs them to the speech recognition parameter control unit 206. For example, the image recognition unit 214 may detect a predetermined keyword from the morphological analysis result and directly convert it as a control parameter such as a speaker, a genre, a language, a dialect, or a situation, or a web service that handles a person name, a place name, or the like May be converted into control parameters using an ontology dictionary, or may be converted into control parameters using an ontology dictionary. In addition, the image recognition unit 214 may designate a temporal range to which the control parameter is applied if the temporal correspondence between the content reproduction time and the control parameter can be acquired. For example, the image recognizing unit 214 can acquire the reproduction time during which text is displayed from metadata or the like. Further, the image recognition unit 214 may convert some or all of the keywords into dictionary parameters. Further, the image recognition unit 214 may convert the recognition result into a language model parameter. The language model parameters can be used for updating (learning) the language model.

  The image recognition unit 214 may switch the conversion method of each word into a control parameter or assign a priority according to the character size, shape (font), screen position, display interval, and the like. The priority can be used to select a valid control parameter when the number of control parameters is excessive. For example, a higher priority is assigned to a larger character size, or a higher priority is assigned if the character shape corresponds to highlighting such as bold, or a specific on-screen position (for example, High priority may be assigned to the bottom of the screen where names are easily displayed. Further, the image recognition unit 214 may recognize a specific mark (logo) representing a specific broadcasting station, program, person, company, organization, product, service, etc. without being limited to characters, and may convert the mark into a control parameter. For example, if the image recognizing unit 214 recognizes a mark indicating a specific program, it may be converted into “control parameter: genre”, “control parameter: language”, etc. corresponding to the program.

  The speech recognition parameter control unit 206 controls speech recognition parameters for speech data to be recognized according to the control parameters from the extracted text processing unit 105 and the image recognition unit 214. For example, the speech recognition parameter control unit 206 roughly selects an acoustic model, a language model, and a word dictionary in accordance with “control parameter: language” or “control parameter: dialect”, and “control parameter: speaker” or “control parameter: genre”. The acoustic model, language model, and word dictionary can be selected in more detail. Further, the voice recognition parameter control unit 206 may select an acoustic model in more detail according to “control parameter: situation”, or may control noise canceling processing. The speech recognition parameter control unit 206 may register the word indicated by the dictionary parameter in the word dictionary, or set the recognition priority of this word high. The speech recognition parameter control unit 206 may update the language model according to the language model parameter. The voice recognition parameter control unit 206 may update the language model or the word dictionary temporarily or continuously.

  The voice recognition parameter control unit 206 may not use some of the input control parameters for controlling the voice recognition parameters. For example, the speech recognition parameter control unit 206 may preferentially use a control parameter from either the extracted text processing unit 105 or the image recognition unit 214, or may use the extracted text processing unit 105 and the image recognition unit 214. Control parameters may be selected without distinction (for example, according to the priority assigned to each control parameter). Further, when the same control parameter is input from both the extracted text processing unit 105 and the image recognition unit 214, the voice recognition parameter control unit 206 may use this control parameter preferentially.

  As described above, the speech recognition apparatus according to the second embodiment controls the speech recognition parameters based on the image recognition result of the video data included in the Web page that provides the content and the content. Therefore, the speech recognition apparatus according to the present embodiment can automatically control speech recognition parameters for content.

(Third embodiment)
As shown in FIG. 3, the speech recognition apparatus according to the third embodiment includes a speech recognition parameter control unit 106 in the speech recognition parameter control unit 306 and a content separation unit 109 in the content recognition unit 309 in the speech recognition apparatus in FIG. And a separated text input unit 315 and a separated text processing unit 316 are added. In the following description, the same parts in FIG. 3 as those in FIG.

  The content separation unit 309 separates audio data and text data included in the media data from the content analysis unit 108. The content separation unit 309 may separate text data included in the metadata. The content separation unit 309 inputs the separated audio data to the audio input unit 110. The content separation unit 309 inputs the separated text data to the separation text input unit 315.

  The separated text input unit 315 converts the separated text data from the content separating unit 309 into a format suitable for the separated text processing unit 316. The separated text input unit 315 inputs the converted separated text data to the separated text processing unit 316.

  The separated text processing unit 316 generates a control parameter based on the separated text from the separated text input unit 315 and inputs the control parameter to the speech recognition parameter control unit 306. Specifically, the separated text processing unit 316 may perform morphological analysis on the separated text. The separated text processing unit 316 converts these morphological analysis results into control parameters and inputs them to the speech recognition parameter control unit 306. For example, the separated text processing unit 316 may detect a predetermined keyword from the morphological analysis result and directly convert it as a control parameter such as a speaker, a genre, a language, a dialect, or a situation, or a Web that handles a person name, a place name, and the like. It may be converted into control parameters using a service, or may be converted into control parameters using an ontology dictionary. Further, the separated text processing unit 316 may specify a temporal range to which the control parameter is applied if the temporal correspondence between the content reproduction time and the control parameter can be acquired. Further, the separated text processing unit 316 may convert some or all of the keywords into dictionary parameters. Further, the separated text processing unit 316 may convert the recognition result into a language model parameter. The language model parameters can be used for updating (learning) the language model.

  The speech recognition parameter control unit 306 controls speech recognition parameters for speech data to be recognized according to control parameters from the extracted text processing unit 105 and the separated text processing unit 316. For example, the speech recognition parameter control unit 306 roughly selects an acoustic model, a language model, and a word dictionary according to “control parameter: language” or “control parameter: dialect”, and “control parameter: speaker” or “control parameter: genre”. The acoustic model, language model, and word dictionary can be selected in more detail. Further, the voice recognition parameter control unit 306 may select an acoustic model in more detail according to “control parameter: situation”, or may control noise canceling processing. The voice recognition parameter control unit 306 may register the word indicated by the dictionary parameter in the word dictionary or set the recognition priority of this word to be high. The speech recognition parameter control unit 306 may update the language model according to the language model parameter. The speech recognition parameter control unit 306 may update the language model or the word dictionary temporarily or continuously.

  The voice recognition parameter control unit 306 may not use a part of the input control parameters for controlling the voice recognition parameters. For example, the speech recognition parameter control unit 306 may preferentially use the control parameter from either the extracted text processing unit 105 or the separated text processing unit 316, or the extracted text processing unit 105 and the separated text processing unit. Control parameters may be selected without distinguishing 316 (eg, according to the priority assigned to each control parameter). In addition, when the same control parameter is input from both the extracted text processing unit 105 and the separated text processing unit 316, the voice recognition parameter control unit 306 may preferentially use the control parameter.

  As described above, the speech recognition apparatus according to the third embodiment controls the speech recognition parameters based on the Web page that provides the content and the text data included in the content. Therefore, the speech recognition apparatus according to the present embodiment can automatically control speech recognition parameters for content.

(Fourth embodiment)
As shown in FIG. 4, the speech recognition apparatus according to the fourth embodiment includes a speech recognition parameter control unit 106 in the speech recognition parameter control unit 406 and a content separation unit 109 in a content separation unit 409 in the speech recognition apparatus in FIG. This corresponds to a configuration in which the video input unit 213 and the image recognition unit 214 in FIG. 2 and the separated text input unit 315 and the separated text processing unit 316 in FIG. 3 are added. In the following description, the same parts in FIG. 4 as those in FIG. 1, FIG. 2, or FIG. 3 are denoted by the same reference numerals, and different parts will be mainly described.

  The content separation unit 409 separates audio data, video data, and text data included in the media data from the content analysis unit 108. Further, the content separation unit 409 may separate text data included in the metadata. The content separation unit 409 inputs the separated audio data to the audio input unit 110. The content separation unit 409 inputs the separated video data to the video input unit 213. The content separation unit 409 inputs the separated text data to the separation text input unit 315.

  The speech recognition parameter control unit 406 controls speech recognition parameters for speech data to be recognized according to control parameters from the extracted text processing unit 105, the image recognition unit 214, and the separated text processing unit 316. For example, the voice recognition parameter control unit 406 roughly selects an acoustic model, a language model, and a word dictionary in accordance with “control parameter: language” or “control parameter: dialect”, and “control parameter: speaker” or “control parameter: genre”. The acoustic model, language model, and word dictionary can be selected in more detail. Further, the voice recognition parameter control unit 406 may select an acoustic model in more detail according to “control parameter: situation”, or may control noise canceling processing. The speech recognition parameter control unit 406 may register the word indicated by the dictionary parameter in the word dictionary, or set the recognition priority of this word high. The speech recognition parameter control unit 406 may update the language model according to the language model parameter. The speech recognition parameter control unit 406 may update the language model or the word dictionary temporarily or continuously.

  The voice recognition parameter control unit 406 may not use part of the input control parameters for controlling the voice recognition parameters. For example, the speech recognition parameter control unit 406 may preferentially use control parameters from some of the extracted text processing unit 105, the image recognition unit 214, and the separated text processing unit 316, or the extracted text processing unit 105, the control parameters may be selected without distinguishing between the image recognition unit 214 and the separated text processing unit 316 (for example, according to the priority assigned to each control parameter). In addition, the speech recognition parameter control unit 406 preferentially uses the control parameter when the same control parameter is input from a plurality of the extracted text processing unit 105, the image recognition unit 214, and the separated text processing unit 316. May be.

  As described above, the speech recognition apparatus according to the fourth embodiment controls the speech recognition parameters based on the Web page that provides the content, the image recognition result of the video data included in the content, and the text data included in the content. To do. Therefore, the speech recognition apparatus according to the present embodiment can automatically control speech recognition parameters for content.

(Fifth embodiment)
As shown in FIG. 5, the speech recognition apparatus according to the fifth embodiment includes a recognition target input unit 501, a content acquisition unit 107, a content analysis unit 108, a content separation unit 509, a speech input unit 510, and a first speech recognition. Unit 517, video input unit 513, image recognition unit 514, separated text input unit 515, separated text processing unit 516, web page acquisition unit 502, web page analysis unit 103, analysis parameter storage unit 104, extracted text processing unit 105, audio It has a recognition parameter control unit 106, a second speech recognition unit 511, and a recognition result output unit 112. In the following description, the same parts in FIG. 5 as those in FIG. 1 are denoted by the same reference numerals, and different parts will be mainly described.

  The recognition target input unit 501 inputs information for acquiring content including audio data to be subjected to voice recognition to the content acquisition unit 107. This information is not limited to the identifier of the Web page that provides the content, but may be address information of a storage medium from which the content is read, a channel through which the content is broadcast, and the like.

  The content separation unit 509 separates audio data, video data, and text data included in the media data from the content analysis unit 108. Further, the content separation unit 509 may separate text data included in the metadata. The content separation unit 509 inputs the separated audio data to the audio input unit 510. The content separation unit 509 inputs the separated video data to the video input unit 513. The content separation unit 509 inputs the separated text data to the separation text input unit 515.

  The voice input unit 510 converts the voice data from the content separation unit 509 into a format suitable for the first voice recognition unit 517 and the second voice recognition unit 511. The voice input unit 510 inputs the converted voice data to the first voice recognition unit 517 and the second voice recognition unit 511. The first voice recognition unit 517 performs voice recognition on the voice data from the voice input unit 510. The first speech recognition unit 517 extracts a word or web page identifier included in the recognition result and inputs the extracted word or web page identifier to the web page acquisition unit 502.

  The video input unit 513 converts the video data from the content separation unit 509 into a format suitable for the image recognition unit 514. The video input unit 513 inputs the converted video data to the image recognition unit 514. Note that the video input unit 513 may thin out frames in the video data from the content separation unit 509 in order to omit image recognition for some frames in the video data.

  The image recognition unit 514 performs image recognition on the video data from the video input unit 513. The image recognition unit 514 extracts a word or Web page identifier from the recognition result and inputs the extracted word or Web page identifier to the Web page acquisition unit 502. Specifically, the image recognition unit 514 recognizes text (for example, a telop, the name of a program performer, a URL of a Web page related to the content, etc.) displayed in the video. The image recognition unit 514 inputs a word or Web page identifier included in the recognition result to the Web page acquisition unit 502.

  In addition, the image recognition unit 514 may assign a priority to a word or Web page identifier according to a character size, a shape (font), a screen position, a display interval, and the like. The priority can be used to select valid word or web page identifiers when the number of word or web page identifiers is excessive. For example, a higher priority is assigned to a larger character size, or a higher priority is assigned if the character shape corresponds to highlighting such as bold, or a specific on-screen position (for example, High priority may be assigned to the bottom of the screen where names are easily displayed. Alternatively, the image recognizing unit 514 recognizes a specific mark (logo) representing a specific broadcasting station, program, person, company, organization, product, service, etc., not limited to characters, and a corresponding word or a corresponding Web page. You may convert into an identifier.

  The separated text input unit 515 converts the separated text data from the content separating unit 509 into a format suitable for the separated text processing unit 516. The separated text input unit 515 inputs the converted separated text data to the separated text processing unit 516.

  The separated text processing unit 516 extracts a word or web page identifier from the separated text from the separated text input unit 515 and inputs the extracted word or web page identifier to the web page acquisition unit 502. Specifically, the separated text processing unit 516 extracts a word or Web page identifier included in the separated text.

  The web page acquisition unit 502 acquires a web page related to the content based on the word or web page identifier from the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516. Specifically, when a word is input, the Web page acquisition unit 502 generates a search expression using this word. The web page acquisition unit 502 transmits this search formula to a predetermined search engine, and acquires a web page from the search result. On the other hand, when a web page identifier is input, the web page acquisition unit 502 acquires the web page according to the web page identifier. The web page acquisition unit 502 inputs the acquired web page to the web page analysis unit 103.

  Further, the Web page acquisition unit 502 may limit the number of words included in the search expression, the number of Web pages acquired from the search result, the number of Web pages acquired according to the Web page identifier, and the like. For example, the web page acquisition unit 502 may preferentially include words from some of the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516, among these, A web page may be acquired by preferentially selecting identifiers of some web pages. Alternatively, the web page acquisition unit 502 weights each word without distinguishing the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516 (for example, according to the priority assigned to each word). A search expression may be generated. Here, weighting means, for example, combining words of a predetermined number or less in descending order of priority, or combining words having a priority of a predetermined value or more. In addition, the Web page acquisition unit 502 receives the same word or the identifier of the same Web page from a plurality of the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516. Alternatively, Web page identifiers may be used preferentially.

  The second voice recognition unit 511 performs voice recognition on the voice data from the voice input unit 510 after the processing of the voice recognition parameter control unit 106 is completed for the voice data to be recognized. The second voice recognition unit 511 inputs the recognition result to the recognition result output unit 112. The first voice recognition unit 517 and the second voice recognition unit 511 may be separate modules or integrated modules.

  As described above, the speech recognition apparatus according to the fifth embodiment is based on the speech recognition result for the audio data separated from the content, the image recognition result for the video data separated from the content, the text separated from the content, and the like. The web page related to the content is acquired, and the speech recognition parameter is controlled based on the related web page. Therefore, according to the speech recognition apparatus according to the present embodiment, it is possible to automatically control speech recognition parameters for content even when there is no Web page providing the content or when it is unknown.

  The speech recognition apparatus according to the present embodiment searches for a web page related to content using audio data, video data, and text data included in the content. However, an effect similar to that of the speech recognition apparatus according to the present embodiment can be obtained without necessarily using all of them. When voice data is not used, the first voice recognition unit 517 in FIG. 5 may be removed. When the video data is not used, the video input unit 513 and the image recognition unit 514 in FIG. 5 may be removed. When text data is not used, the separated text input unit 515 and the separated text processing unit 516 may be removed.

(Sixth embodiment)
As shown in FIG. 6, the speech recognition apparatus according to the sixth embodiment replaces the recognition target input unit 501 in FIG. 5 with the recognition target input unit 601 and the web page acquisition unit 102 with the web page acquisition unit 602. Corresponds to the configuration. In the following description, the same parts in FIG. 6 as those in FIG.

  The recognition target input unit 601 inputs, to the Web page acquisition unit 602 and the content acquisition unit 107, an identifier of a Web page that provides content including audio data that is a target of voice recognition.

  The web page acquisition unit 602 acquires a web page according to the web page identifier from the recognition target input unit 601. Similarly to the Web page acquisition unit 502, the Web page acquisition unit 602 relates to content based on words or Web page identifiers from the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516. Web page to be acquired is acquired. The web page acquisition unit 602 inputs the acquired web page to the web page analysis unit 103.

  As an example, the Web page acquisition unit 602 first acquires a Web page according to the Web page identifier from the recognition target input unit 601. If the extracted text processing unit 105 cannot obtain sufficient control parameters for the Web page, the web page acquisition unit 602 uses words from the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516. Or you may acquire additionally the web page relevant to a content based on the identifier of a web page.

  As another example, the web page acquisition unit 602 first creates a web page related to content based on a word or web page identifier from the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516. To get. If the extracted text processing unit 105 cannot obtain sufficient control parameters for the Web page, the Web page acquisition unit 602 additionally acquires the Web page according to the Web page identifier from the recognition target input unit 601. Also good.

  As another example, the web page acquisition unit 602 includes a web page according to the web page identifier from the recognition target input unit 601, a word from the first speech recognition unit 517, the image recognition unit 514, and the separated text processing unit 516. You may acquire in parallel with both the web page relevant to the content based on the identifier of a web page.

  As described above, the speech recognition apparatus according to the sixth embodiment controls the speech recognition parameters based on at least one of a web page that provides content and a web page related to the content. Therefore, according to the speech recognition apparatus according to the present embodiment, speech recognition parameters can be automatically controlled.

  The speech recognition apparatus according to the present embodiment searches for a web page related to content using audio data, video data, and text data included in the content. However, an effect similar to that of the speech recognition apparatus according to the present embodiment can be obtained without necessarily using all of them. When voice data is not used, the first voice recognition unit 517 in FIG. 6 may be removed. When video data is not used, the video input unit 513 and the image recognition unit 514 in FIG. 6 may be removed. When text data is not used, the separated text input unit 515 and the separated text processing unit 516 may be removed.

  Note that the present invention is not limited to the above-described embodiments as they are, and can be embodied by modifying the components without departing from the scope of the invention in the implementation stage. Various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments. Further, for example, a configuration in which some components are deleted from all the components shown in each embodiment is also conceivable. Furthermore, you may combine suitably the component described in different embodiment.

  For example, it is possible to provide a program that realizes the processing of each of the above embodiments by storing it in a computer-readable storage medium. The storage medium may be a computer-readable storage medium such as a magnetic disk, optical disk (CD-ROM, CD-R, DVD, etc.), magneto-optical disk (MO, etc.), semiconductor memory, etc. For example, the storage format may be any form.

  Further, the program for realizing the processing of each of the above embodiments may be stored on a computer (server) connected to a network such as the Internet and downloaded to the computer (client) via the network.

DESCRIPTION OF SYMBOLS 101 ... Recognition object input part 102 ... Web page acquisition part 103 ... Web page analysis part 104 ... Analysis parameter memory | storage part 105 ... Extraction text processing part 106 ... Speech recognition parameter control part 107 ... Content acquisition unit 108 ... Content analysis unit 109 ... Content separation unit 110 ... Voice input unit 111 ... Voice recognition unit 112 ... Recognition result output unit 206 ... Voice recognition parameter control 209 ... Content separation unit 213 ... Video input unit 214 ... Image recognition unit 306 ... Speech recognition parameter control unit 309 ... Content separation unit 315 ... Separated text input unit 316 ... Separate text processing unit 406... Speech recognition parameter control unit 409... Content separation unit 501. Elephant input unit 502 ... Web page acquisition unit 509 ... Content separation unit 510 ... Audio input unit 511 ... Second audio recognition unit 513 ... Video input unit 514 ... Image recognition unit 515 ... Separated text input unit 516 ... Separated text processing unit 517 ... First speech recognition unit 601 ... Recognition target input unit 602 ... Web page acquisition unit

Claims (11)

A content acquisition unit for acquiring content including audio data;
A web page acquisition unit for acquiring a web page providing the content;
A web page analysis unit that performs analysis based on a web page that provides the content and extracts text indicating characteristics of the audio data;
A parameter control unit for controlling speech recognition parameters for the speech data based on the extracted text;
A speech recognition apparatus comprising: a speech recognition unit that performs speech recognition on the speech data in accordance with a controlled speech recognition parameter. A storage unit that associates and stores the identifier of the Web page and the analysis target and extraction method for extracting the text;
The web page analysis unit extracts the text from an analysis target corresponding to an identifier of a web page providing the content according to an extraction method corresponding to the identifier of the web page providing the content.
The speech recognition apparatus according to claim 1. A storage unit that stores the identifier of the Web page, the narrowing condition, and the application target of the narrowing condition in association with each other;
The Web page analysis unit extracts the text if the narrowing condition corresponding to the identifier of the Web page providing the content is not satisfied in the application target of the narrowing condition corresponding to the identifier of the Web page providing the content. Is omitted,
The speech recognition apparatus according to claim 1. The storage unit further stores a processing method for converting the extracted text into a control parameter in association with the identifier of the Web page,
The control unit controls the speech recognition parameter according to a control parameter in which the extracted text is converted according to a processing method corresponding to an identifier of a Web page that provides the content.
The speech recognition apparatus according to claim 2.   If the Web page providing the content is a predetermined Web page, the Web page analyzing unit searches for another Web page using the extracted text, and determines a predetermined Web page corresponding to the identifier of the other Web page. The speech recognition apparatus according to claim 1, wherein the text is extracted from an analysis target according to a predetermined extraction method.   The speech recognition apparatus according to claim 1, wherein the parameter control unit controls the speech recognition parameter based on a description position of the text.   The speech recognition apparatus according to claim 1, wherein the parameter control unit controls the speech recognition parameter based on the predetermined keyword if the text includes a predetermined keyword. A content acquisition unit for acquiring content including audio data;
A web page acquisition unit that acquires a web page related to the content based on at least one of a speech recognition result of the audio data, an image recognition result of video data separated from the content, and text data separated from the content When,
A web page analysis unit that performs analysis based on a web page related to the content and extracts text indicating characteristics of the audio data;
A parameter control unit for controlling speech recognition parameters for the speech data based on the extracted text;
A speech recognition apparatus comprising: a speech recognition unit that performs speech recognition on the speech data in accordance with a controlled speech recognition parameter.   The web page acquisition unit weights at least one of a first word included in the speech recognition result, a second word included in the image recognition result, and a third word included in the text data. The speech recognition apparatus according to claim 8, wherein a search expression for searching for a Web page related to the content is generated.   The speech recognition apparatus according to claim 9, wherein the Web page acquisition unit performs weighting on the second word based on a character size, a shape, or an in-screen position.   9. The voice recognition according to claim 8, wherein if the image recognition result matches a predetermined mark, the Web page acquisition unit acquires a Web page related to the content according to a predetermined identifier corresponding to the predetermined mark. apparatus.

Images