WO2008044669A1 - Audio information search program and its recording medium, audio information search system, and audio information search method - Google Patents

Abstract

[PROBLEMS] To provide an audio information search program which can obtain an accurate search result and can be easily used in various databases. [MEANS FOR SOLVING PROBLEMS] An audio information search program causes a computer to execute at least: a step of converting an audio element recorded in audio information into a character string, a step of measuring a reproduction time of the audio information, a step of registering the character string of the audio information in a database by dividing the string into a predetermined time interval if the reproduction time of the audio information is longer than the predetermined time and registering the character string of the audio information in the database without dividing it if the reproduction time of the audio information is not longer than the predetermined time, a step of deciding a search item for searching the database and a search condition for the search item, a step of executing a search of the database according to the search condition, and a step of outputting the search result.

Description

 Specification

 Voice information search program and its recording medium, voice information search system, and voice information search method

 Technical field

 [0001] The present invention relates to a speech information retrieval program. More specifically, the present invention relates to a voice information search program and its recording medium, a voice information search system, and a voice information search method. Background art

 With the recent promotion of computerization, various electronic information such as text information, image information, audio information, and moving image information has been converted into data. Then, a user can access a database in which such information is stored, and obtain a desired information file by searching in the database. In this regard, as the types of information files registered in the database to be searched are diversified, search methods have been developed.

 [0003] For example, by inputting a keyword to an audio information file such as a lecture or a news program, searching for an audio information file including the keyword is performed! In that case, it is important how efficiently and accurately the time zone (playback time position) in which the desired keyword is issued in the audio information file can be searched. For example, in a movie file or audio information file for several hours of movie or speech, it is possible to efficiently search for the scene that the user wants to see (the scene where the desired speech is emitted and its playback time position). The stress when using search is reduced by reducing the user's time burden.

[0004] As such a speech information retrieval method, there is a method of converting a search keyword input by a user into a speech element and performing matching with speech stored in a speech information file of a database. . Patent Document 1 describes a technique for searching for speech information in consideration of the possibility that a recognition error has occurred when generating a speech element sequence from an input arbitrary word. It ’s been disclosed.

Patent Document 1: Japanese Patent Application Laid-Open No. 2005-257954. Disclosure of the invention Problems to be solved by the invention

 [0006] However, when searching for voice information, if the voice information file stored in the database is used as a voice element, the voice may not be recognized correctly due to the intonation of the voice, the presence or absence of strength, or the like. , Search accuracy decreases. Also, if the amount of information recorded in the audio information file (recording time, recording time, etc.) is large, the desired scene (playback time position) cannot be searched from this audio information file. In some cases, search accuracy decreases.

 [0007] Therefore, the main object of the present invention is to provide a speech information search program that can obtain more accurate search results and can be easily used in various databases. Means for solving the problem

 [0008] First, as a first aspect of the present invention, a step of converting an audio element recorded in audio information into a character string, a step of measuring the reproduction time of the audio information, and a reproduction time of the audio information If it is longer than a predetermined time, the character string of the audio information is divided into predetermined time intervals and registered in the database. If the reproduction time of the audio information is less than the predetermined time, the character of the audio information is Registering in the database without separating columns, determining a search item for searching the database, and a search condition for the search item, and for the database according to the search condition There is provided a voice information search program for causing a computer to execute at least a step of executing a search and a step of outputting a result of the search. By causing the computer to execute each of these steps, it is possible to perform voice information search with high search accuracy and high search speed.

 [0009] In addition, as a second aspect of the present invention, there is a step of measuring the number of characters constituting the character string, segmenting if the number of characters is longer than a predetermined number of characters, and registering it directly in the database if it is not longer You can be like fi.

 [0010] Next, in the present invention, in order to perform a search on the database, at least

 (1) A step of combining two pieces of audio information before and after the divided audio information,

(2) a step of determining whether the character string specified by the search condition is included in the character string of the two combined voice information, or (3) each sentence of the two separated voice information. There is provided a speech information search program characterized by causing a computer to execute a step of determining whether a character string specified by a search condition is included in a character string of each speech information. By separating speech information, it is possible to perform speech information search with high search accuracy even for character strings that are truncated or truncated.

Subsequently, in the present invention, in order to output the search result, at least (a) a recording time position of voice information in which a voice element corresponding to the character specified in the search condition is recorded is specified. (B) a step of designating a recording time position that is a predetermined time before the designated recording time position as a reproduction start position, and (c) starting from the designated reproduction start position. Provided is a voice information retrieval program characterized by causing a computer to execute a voice reproduction step. By causing the computer to execute each of these steps, the sound S can be reproduced from the more accurate reproduction position that matches the search condition.

 Furthermore, the present invention provides a computer-readable recording medium on which the program is recorded. The “recording medium” used in the present invention means any computer-readable recording medium on which a program is recorded, which is used for program installation, execution, program distribution and distribution, and the like.

 [0013] Also, in the present invention, the voice element recorded in the voice information is converted into a character string, the voice information playback time measuring means, and the voice information playback time is longer than a predetermined time. The character string IJ of the voice information is registered in the database by dividing it at predetermined time intervals, and when the reproduction time of the voice information is equal to or shorter than the predetermined time, the character string of the voice information is not divided into the database. Means for registering in the database, search items for searching the database, means for determining search conditions for the search items, means for executing a search for the database according to the search conditions, and the search And a speech information retrieval system comprising at least a result output means. By using such a voice information search system, it is possible to construct a voice information search system having high search accuracy and a fast search speed V.

[0014] In the present invention, a procedure for converting an audio element recorded in audio information into a character string, a procedure for measuring a reproduction time of the audio information, and a reproduction time of the audio information from a predetermined time If the length of the voice information is too long, the character 歹 | J of the voice information is divided into predetermined intervals and registered in the database. If the playback time of the voice information is less than the predetermined time, the voice information A procedure for registering in the database without dividing the character string, a search item for searching the database, a procedure for determining a search condition for the search item, and a search condition for the search item, and the search condition Accordingly, there is provided a voice information search method comprising at least a procedure for executing a search on the database and a procedure for outputting a result of the search. According to the voice information retrieval method, it is possible to perform voice information retrieval with high retrieval accuracy and high search speed.

It should be noted that “audio information” in the present invention means any information in which at least an audio element is recorded, unless otherwise specified, and includes, for example, moving image information. Similarly, the “voice information file” in the present invention includes at least a moving picture information file in which a voice element is recorded, unless otherwise specified. In addition, steps, procedures, means, etc. in the present invention are not limited to the order of performing these steps as long as the object of the present invention can be achieved, and within the scope of the present invention! ,.

 The invention's effect

 [0016] According to the audio information search program of the present invention, it is possible to search audio information with high search accuracy and to perform audio information search with a high search speed.

 BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that the embodiments described below exemplify preferred embodiments of the present invention, and the present invention is not construed to be narrowly limited thereby. This application claims priority based on Japanese Patent Application No. 2006-277026 filed on Oct. 10, 2006. The contents described in the specification, drawings, etc. of the patent application are It is built in.

FIG. 1 is a conceptual diagram for explaining a speech information retrieval program according to the present invention.

As shown in FIG. 1, a speech information retrieval program according to an embodiment of the present invention includes a speech information file text conversion function 1, a speech information file playback time measurement function 2, and a database registration function 3. Search condition determination function 4, search processing function 5, and search result output function 6 are executed at least by the computer. 7 is accumulated. Each function will be described below.

 [0020] The text conversion function 1 of the voice information file is a function of converting a voice element input to the voice information file into a character string. Audio information file playback time measurement function 2 is a function for measuring the playback time of an audio information file. The database registration function 3 is a function for registering the voice information file in the database 7. The search condition determination function 4 is a function for determining a search condition for extracting a desired audio information file. The search processing function 5 is a function for executing a search on the database 7 in accordance with the search conditions determined by the search condition determination function 4. The search result output function 6 is a function for outputting the search result obtained by the search processing function 5.

 FIG. 2 is a flowchart when registering a database in the first embodiment of the speech information retrieval program according to the present invention. That is, FIG. 2 is a flowchart for registering an audio information file in the database based on the playback time.

 [0022] First, file information of an audio information file (including a moving image information file; the same applies hereinafter) is input (Sl). As the file information, the title name, category, creator, voice language, creation date, etc. of the voice information file are input. Thus, in the present invention, when content information is registered as audio information, the ability to input filter condition items such as content name and category is not particularly limited, and the file information input in this step is not particularly limited. Information necessary for database construction can be input. For example, basic content information may be input. The input method is not particularly limited in the present invention, and can be performed by, for example, a keyboard. Various information input in this way is registered in the database as file information of the audio information file (S2).

 [0023] Next, it is determined whether the sound element of the sound information file is Japanese or English (S3). As a result, if it is determined that the audio information file is in Japanese, noise cut processing for Japanese is performed (S4a). If it is determined that the sound information file is in English, noise cut processing is performed for English! / S (b4). This noise cut should be processed while playing back the audio information file.

As described above, in the present invention, an audio information file is reproduced, converted into text using an audio conversion system, and stored in a database. When the played audio is in Japanese (S 3, see S4a), and the text conversion is displayed in hiragana. Or, it is not particularly limited to display Japanese translation! / Katakana display.

 [0025] When the reproduced voice is in a foreign language such as English, the text conversion is displayed in the foreign language such as English characters. In the present invention, the target language is not particularly limited, and may be Japanese, English, Chinese, Korean, French, etc. In this case, it is preferable to use a language that does not have speech. In the case of speech information, the words (characters) corresponding to the speech can be uniquely converted and determined, so that higher search accuracy can be obtained. .

 [0026] In the present invention, steps S1 to S4 are appropriately performed from the viewpoint of search items used as search conditions that are not necessarily required, the viewpoint of easy management of audio information files on a database, and the like. Can do.

 Next, the audio information file is reproduced (S 5), and the audio elements of the reproduced audio information file are converted to text (S 6). In the present invention, the method for text-converting a voice element of a voice information file is not particularly limited. For example, text conversion can be performed using a voice conversion system such as Microsoft (registered trademark) R.Net Speech.

 [0028] Then, it is determined whether the reproduction time of the reproduced audio information file is longer than n seconds! / Or shorter! / (S7). As a result, when it is determined that the time is shorter than n seconds, the text information of the voice information file is registered in the database (S 8a). The audio information file processed in step S8a is a short-time audio information file with a playback time of n seconds or less.

[0029] If it is determined in step S7 that it is longer than n seconds, the playback part of the first n seconds of the audio information file is separated, and the text information of the separated audio information file is separated. Register in the database (S8b). Then, the remaining portion of the cut audio information file is determined again in step S7 as to whether or not the reproduction time is longer than n seconds, and is finally divided in units of audio information file power seconds. The audio information file processed in step S8b is an audio information file whose playback time is longer than the predetermined time n seconds.

That is, in the present invention, the reproduced audio information is converted into text and stored in the database. At that time, each time stamp is pressed at predetermined time intervals (n seconds), Recorded in units. Also, if the audio information is short and / or information amount less than the predetermined time interval (n seconds), the time stamp is recorded without being divided (see FIG. 3 etc.).

 [0031] In the present invention, the time setting of the predetermined time n seconds in step S7 is not particularly limited, and can be set as appropriate in consideration of the search accuracy, the processing capability of the computer or the like used, the usage environment, and the like. For example, if you want to increase the search accuracy, you can shorten the n-second time setting. If you want to focus on the search speed, etc., you can increase the n-second time setting.

 [0032] As described above, all audio information files that have finally reached the predetermined time n seconds or less are registered in the database.

 FIG. 3 is a conceptual diagram for explaining the basic information table of the audio information file registered in the database in the first embodiment (see FIG. 2, S1, etc.), and FIG. 4 is the first embodiment. FIG. 3 is a conceptual diagram illustrating content data of an audio information file registered in a database in the form (see FIG. 2; S7, S8a, S8b, etc.).

 FIG. 3 is an example of the file information of the audio information file, and four programs with content IDs “1” to “4” are registered. Information about the titles, categories, and audio languages of the four programs is shown. For example, the content ID “1” is assigned to the first registered program (“Create homepage”) as the number that identifies and identifies the program. “Computer” is assigned to the power category. Thus, the search accuracy can be further improved by registering the genre and the like of the content program. Then, “Japanese” as the voice language is registered as file information.

[0035] FIG. 4 shows the content data of the audio information file registered in the database, and the audio information file is converted to text and divided so as to have a reproduction time of a predetermined time n seconds or less ( Figure 2; see S8a, 8b etc.). FIG. 4 illustrates the program “homepage creation” with the content ID “1” in FIG. 3 and the program “weather in Tokyo today” with the content ID “4”. . The content ID is a number assigned to each program as in FIG. The record ID indicates the power of the audio information played back in the program. For example, record ID “1” is the first audio information file played back in the program. Record ID “2” is played after record ID “1”. Show that it is the second audio information file! /

 [0036] The time stamp in Fig. 4 indicates the playback time from the beginning of the program. For example, the time stamp “00:00:00” indicates that it is 0 seconds from the beginning of the program, and the time stamp “00:00:00” is exactly 10 seconds from the beginning of the program. It is shown that it was time. That is, in this case, the predetermined time n of each record is 10 seconds (see FIG. 2; S8b, etc.).

 On the other hand, in FIG. 4, for the program “Today's Tokyo Weather” with the content ID “4”, the time stamp is “00:00:00” and the record ID is “1” only. This is a short program with a total playback time of 8 seconds (see Fig. 2, S8a). Therefore, this program is an audio information file shorter than the predetermined time n (n = 10 seconds), and the record “ID” consists only of “1”.

In addition, the item “text” in FIG. 4 indicates data obtained by converting the voice element of the voice information file into characters. For example, in the record ID “1” of the content ID “1”, the voice of “I will make a home page from now on! The last “Sor…” is divided into a predetermined time n (n = 10 seconds). The next record ID “2” is a continuation of the record ID “1”, and the text of the record ID “2” must be “····. Muriyo “•” audio is converted to text!

 FIG. 5 is a flowchart for performing a search process for the database in the first embodiment. That is, FIG. 5 is a flowchart for performing a search process for the database registered in FIG. 2, and shows a search process for searching for an audio information file including a predetermined keyword.

 First, a search condition is input (S9). First, the search condition items to be entered include the category, language, text keyword, etc. of the voice information file. In the present invention, items of search condition items (see S2 in FIG. 2) input at the time of database creation can be input. In the present invention, the search condition items are not particularly limited. For example, a category, a language, a keyword included (character 歹 IJ), and the like can be set.

[0041] Next, based on the registered content basic information table, the set category and language are set. Search and extract data having basic content information that meets the conditions such as words (S10). This step is based on the basic information (see Figure 2, Figure 3, etc.) given to the audio information file.

 [0042] As a result of step S10, it is determined whether the number of extractions is 0 or more than 1 (Sl l). As a result, if there are more extractions that match the search conditions, a content ID list of the content ID (see Fig. 2 etc.) of the corresponding data is created. On the other hand, if the number of extracted cases is 0, “N / A” is displayed (12b).

 [0043] Subsequently, the number of content ID lists is looped (S13 to S26). That is, all the contents extracted in step S 10 are searched. For example, when a search is made in the database using the category “computer” as a search condition, all the contents belonging to the category “computer” are extracted, and all the records of the content group are searched for text.

 [0044] The text search is performed in the following steps. First, the content data count is reset (S14), and a search is made for records with matching content IDs in the same content data table (S15 to 25). First, a search is made for a group of records that are divided and registered within the content (within the program). For the record ID, the nth ID and the (n + 1) th ID are extracted (S16). For example, records with record IDs “1” and “2” are extracted. Alternatively, records with record IDs “2” and “3” are extracted.

Next, the texts of the extracted record IDs “n” and “n + 1” are combined (S17). Then, a keyword search is performed on the character strings in the combined record IDs “n” and “n + 1”. In this way, by combining the two preceding and following record IDs and performing a text search, you can also search for the delimited words before and after the record ID (cut off the head and the bottom). For example, in the example of Figure 3, by combining the content IDs “2” and “3”, the word “Muryo / U (free)” separated by the content IDs “2” and “3” is also searched. can do. In other words, from the beginning of content ID “2” to the last character of content ID “3” J IJ “Because it is free, it ’s free, so you can make it.” (See Figure 3) Do. As a result, if the corresponding keyword is included, the process proceeds to step S20, and if there is no corresponding record, the process proceeds to step 26. In step S20, keyword search is performed for the character string in the text of the content “n”. In this step S20, text search is performed within a single record without combining previous and subsequent records. For example, in the example of FIG. 3, a keyword search is performed on the character string in the content ID “2”.

 As a result, if the corresponding character string is not extracted, a keyword search is also performed for the character string in the text of the subsequent record ID “n + 1” (S22a). On the other hand, when the corresponding character string (ie, search keyword) is extracted in the single record “n”, the content D information and record D information of the record ID “n” are added to the search list. (S22b). That is, in this case, a search keyword exists in the record ID “n”.

 [0048] If the corresponding character string (ie, search keyword) is not extracted even in the single record “n + 1”, the content ID information of both the record IDs “n” and “n + 1” is used. And record ID information are added to the search list (S24a). That is, in this case, the search keyword exists across the record IDs “n” and “n + 1”.

 [0049] On the other hand, when the corresponding character string (ie, search keyword) is extracted in the single record “n + 1”, the content D information and record D information of the record ID “n + 1” are retrieved from the search list. (S24b). In other words, in this case, the search keyword exists in the record ID “n + 1”.

 Subsequently, the same content IDs “n + 1” and “n + 2” are searched similarly (S15 to S25). A search is performed for all record IDs belonging to one content.

 [0051] Next, from the search result list obtained so far, data in which the corresponding content ID and the same item in the content basic information table match is acquired (S28). Then, the acquired data is converted into display data (S29). In the present invention, the aspect of the display data is not particularly limited, but may be an aspect listed on the user terminal, for example.

[0052] Subsequently, data in which the corresponding content ID matches the same item in the content basic information table is acquired (S30). Then, the acquired data is converted into display data (S31). In the present invention, the mode of the display data is not particularly limited, but may be a mode listed in one user terminal, for example. Then, repeat steps S28 to S31 as many times as there are records in the search list (S27 to 32). FIG. 6 is a flowchart when outputting the detection result in the first embodiment. That is, FIG. 6 is a flowchart for outputting the search result obtained by the search processing in FIG. 5, and is a search result output for reproduction from a reproduction portion including a predetermined key keyword.

 First, it is selected whether or not to output the playback position in the accuracy improvement mode (S33). When the accuracy improvement mode is selected, more accurate keyword playback is possible. When the accuracy improvement mode is selected, the start time position of the corresponding content data is first obtained from the result list obtained in the search process (see FIG. 4 etc.) (S34). Subsequently, the output audio information file is selected from Japanese or English (S35). In the present invention, the speech language is not limited to Japanese and English, and may be a plurality of types of languages.

 In the case of Japanese, noise cut processing based on Japanese speech is performed (S36a).

 In the case of English, noise cut processing based on English speech is performed (S36b). Since the utterance wave number varies depending on the language, it can be reproduced with little noise when it is reproduced and output by performing an appropriate noise cut process according to the language. This noise cut should be processed while playing back the audio information file.

 Next, information on the start time position of the corresponding content data is acquired from the search result list obtained in the search process (see FIG. 4 etc.) (S36). Through this step, information on the scheduled playback position of which audio ID file (that is, which content ID) should be played back from what record ID of what record ID is acquired. Then, according to the information of the scheduled playback position, playback is started from the corresponding scheduled playback position (S37), and the playback location is converted to text (S38). Once the voice information of voice information is converted to text and verified whether the text really contains the key words, the search accuracy can be improved.

[0057] Then, the first character 1S of the keyword 1S is determined as to the power contained in the text at the scheduled playback position (S39). For example, when a voice information file including the keyword “Internet” is searched, it is a step of determining whether or not the first character “I” of “Internet” is included in the scheduled playback position. As a result, if the first character is included, the playback position obtained by subtracting a predetermined time (1 second) from the playback start time (scheduled playback position) of the first character is stored in a variable (S40). [0058] In this step S40, for example, in the case where the keyword "Internet" is searched for the content "homepage creation" in FIG. ”And the playback start time at which the first character“ i ”of the keyword“ Internet ”is located is 35 seconds after the beginning of the content, the playback position is 34 seconds one second before that position. Is stored in a variable. As a result, it is possible to prevent the voice information finale from being played when the first character of “Internet” is cut off and the head is cut off like “Internet”, and the sound is played back right before “Internet”. be able to.

 In the present invention, the predetermined time may not be 1 second, but can be determined appropriately according to the type of search target file, the audio content, and the like.

 [0060] If it is determined in step S39 that the first character of the keyword does not match, the voice information file at the scheduled start position is played again (S37), and the second character (n It is determined whether or not 2) matches (S44).

 [0061] In this way, the same steps are performed for the second and subsequent characters of the keyword, and it is determined whether the keyword word (or phrase or sentence) is included in the text at the playback planned position (S4;! To S45). ). That is, for the determination after the second character of the keyword, the voice information file at the scheduled start position is also reproduced (S42), and this is converted to text (S43). Then, it is determined whether the converted text matches the nth character (n = 3, 4, ···) of the keyword (S44). This loop of steps S4;! To S45 is repeated until all n characters of the keyword match.

 [0062] Then, the start time position of the corresponding content data in the search result list and the time confirmed in the variable are calculated (S46). For example, in the case of Figure 3, if the playback start position of the keyword “Internet” is 35 seconds after the beginning of the content, the position of 34 seconds is calculated by subtracting the predetermined time of 1 second from 35 seconds. To do.

 In this way, the reproduction start position calculated in step S46 is set (designated) (S47), and reproduction is performed from the reproduction designated position of the predetermined audio information file (S48).

[0064] If the accuracy improvement mode is not selected in step S33, the start time position of the corresponding content data is directly acquired from the result list obtained by the search process. (S10b), the playback is started from the start time position (S47). For example, if the content data “Homepage creation” in FIG. 3 is searched with the keyword “Homepage” and then the keyword “Homepage” is searched, the first playback start position of the record ID “4” containing the keyword (ie, the content Playback starts from 30 seconds from the beginning).

That is, in the present invention, by providing the accuracy improvement mode, the accuracy improvement mode can be appropriately set to “valid” or “invalid”. To enable the accuracy improvement mode, set the accuracy improvement mode to “Enable”. As a result, playback is performed from the time stamp of the record to be searched, and the location matching the keyword is specified. Then, the total time of the time until matching and the time of the time stamp is set as the playback start position for the user.

 [0066] If the accuracy improvement mode is not desired to function, the accuracy improvement mode is set to "invalid". As a result, playback starts from the time stamp of the record to be searched (that is, from the beginning of the record ID to be searched).

 FIG. 7 is a flowchart showing a procedure for registering a database in the second embodiment of the speech information search program according to the present invention. In the second embodiment, the audio information file according to the present invention is registered in a database based on a predetermined number of characters. The following description will focus on differences from the first embodiment of the present invention.

 [0068] First, file information of an audio file is input (S1), and the audio file is converted to text.

 The procedure up to (S6) is the same as in FIG. It is determined whether the number of characters in the voice information file converted into text in this way is larger or smaller than a predetermined number of characters (n in FIG. 7) (S7-2).

 [0069] As a result, when it is determined that there are fewer than n characters! /, The text information of the voice information file is registered in the database (S8a). That is, the audio information file processed in step S8a is an audio information file composed of short // text with a reproduction time of n characters or less.

[0070] If it is determined in step S7-2 that the number of characters is longer than n, the first n number of playback parts of the audio information file are separated, and the audio information file text is delimited. Information is registered in the database (S8b). And the cut audio information For the rest of the file, it is determined again in step S7-2 whether the text is longer than n characters. Finally, the audio information file is divided in units of n characters. That is, the audio information file processed in step S8b is an audio information file in which the number of texts in the reproduction information is longer than the predetermined number of characters (n characters).

 That is, in the present invention, the reproduced audio information is converted into text and stored in the database. At this time, by pressing a time stamp with a predetermined number of characters (n characters), each unit unit It is recorded separated by. If the audio information has a short amount of information equal to or less than the predetermined number of characters (n characters), the time stamp is pressed and recorded as it is without being divided.

 [0072] As described above, by dividing the audio information by a predetermined number of characters, it is possible to cut the unvoiced part, the music-only part, etc. in the audio information. In other words, since no records are created for silent parts, data enlargement can be prevented. Therefore, even for long-time files, it is possible to store text data (data obtained by converting voice information into text) with a minimum data capacity. This method is particularly effective for long-time audio files that contain many parts other than lyrics and lines, such as music, movies, and musicals.

Yes

 [0073] Then, after the audio information file is registered in the database, search processing, search result output processing, and the like can be performed in the same procedure as in the first embodiment of the present invention (Figs. 5 and 5). (Refer to 6).

 [0074] FIG. 8 is a flowchart showing a procedure for registering a database in the third embodiment of the speech information search program according to the present invention. In the third embodiment, the audio information file according to the present invention is registered in the database based on the predetermined number of words. The following description will focus on differences from the first embodiment and the second embodiment of the present invention.

 [0075] First, the procedure from inputting the file information of the audio information file (S1) to text conversion of the audio information file (S6) is the same as in Figs. The voice information file converted into text in this way is divided into phrases or words as a unit (S73).

That is, in this embodiment, the text is divided into predetermined phrase units or word units. The delimiter unit may be a phrase unit or a word unit, and can be selected as appropriate. In the present invention, the method for recognizing the phrase unit or the word unit is not particularly limited, and a suitable method can be used as appropriate. Also, the unit to be divided is not particularly limited. For example, although it is not shown, a predetermined phrase unit or word unit) can be set as n clause number or n word number). If the unit is shortened, the search accuracy is improved, and if the unit is lengthened, the search speed can be increased. A suitable unit can be appropriately determined in consideration of these factors.

[0077] As a result, if the number of phrases or the number of words of the voice information file subjected to text conversion is less than the predetermined number of n phrases (or n words), it is registered in the database at the end of playback and processed. The process ends (S8a). That is, the voice information file processed in step S8a is a voice information file consisting of short text and text whose playback time is less than the number of possessed phrases n (predetermined number of words n).

 [0078] If it is determined in step S7-3 that it is longer than n clauses (n words), the playback portion of the first n clauses (n words) of the audio information file is selected. Divide and register the text information of the separated audio information file in the database (S8b). Then, the remaining portion of the cut audio information file is determined again in step S 7-3 whether or not the text is longer than n clauses (n words). Finale is separated by n clauses (n words). That is, the audio information file processed in step S8b is an audio information file in which the number of texts of the reproduction information is longer than the predetermined number of phrases (predetermined number of words).

 [0079] FIG. 9 is a flowchart for performing a search process for a database according to the third embodiment. That is, FIG. 9 is a flowchart for performing a search process for the database registered in FIG. The difference from the case where the audio information file is registered in the database based on a predetermined time (see FIG. 6 etc.) will be described below.

 [0080] First, the search condition is input (S9) until the procedure (S10, S11) for extracting data from the content basic information table from a predetermined condition such as category 'language', the first embodiment of the present invention, The same as in the second embodiment.

[0081] Then, if there are more than the number of extracted cases in step S11, the corresponding data is copied. A keyword in the text of the content data table is searched from the content ID (S49 a) _{D, that} is, a step of further searching the content ID.

 [0082] If the number of extractions in the search result of step S11 is 0, "N / A" is displayed and the search is terminated (S49b).

 [0083] If there is one or more extraction results as a result of keyword search in step S49a, a loop search is performed for the corresponding number (S51a to S57). In step S49a, if the number of extractions is 0, “N / A” is displayed and the search ends (S49b).

 [0084] When the number of extractions is one or more, the following loop search is performed (S51a to S57). First, the content ID and record ID “n” are added to the result list. In other words, this step identifies the record ID that includes the corresponding audio part. Further, the data corresponding to the same item in the content ID and the content basic information table is acquired from the result list (S53). Subsequently, the acquired data is converted into display data (S54). For example, it may be display data that can be displayed in a list on the user's terminal device and the corresponding part can be reproduced as it is by clicking on the linked part.

 [0085] A loop search is performed (S57) in which the series of steps (S52 to S56) are all performed for the corresponding number of cases.

 Thus, as in the third embodiment according to the present invention, by separating audio information based on the number of phrases and the number of words, adjacent record IDs “n” and record IDs “n + 1” are obtained. High-accuracy voice information retrieval can be performed without performing a process of combining and retrieving (for example, see S15 to S25 in FIG. 5). In other words, since it is registered in the database in terms of phrases or words that make sense in advance, the record ID ends in a truncated state (for example, the record ID “2” in the content ID “1” in FIG. 4). Reference) can be prevented.

[0087] In this way, by performing registration and processing based on the number of phrases or the number of words as in this embodiment, it is possible to register in the database by omitting parts such as unvoiced parts in the audio file, It is not necessary to search by combining record IDs of text data! /, Etc., so that the burden on the server can be reduced and higher speed search is possible. FIG. 10 is a flowchart for outputting the detection result in the third embodiment. That is, FIG. 10 is a flowchart for outputting the search result obtained by the search process in FIG. The difference from the case where the audio information file is registered in the database based on a predetermined time (see FIG. 7 etc.) will be described below.

 [0089] The start time position of the corresponding content data is acquired from the result list obtained by the search processing performed in FIG. 9 (S58). Subsequently, the start time of the content data is set (S59). Then, playback is performed from the set start time (S60).

 [0090] Since registration and processing are performed based on the number of phrases or the number of words) as in this embodiment, data can be registered on a phrase basis or word basis. There is no need to set or select the accuracy improvement mode (see S33 in Fig. 6) during search as in the first and second embodiments. Therefore, faster search processing can be performed, the burden on the server can be reduced, and a search with high search accuracy similar to the accuracy improvement mode can be performed. Furthermore, it can handle a wide variety of audio information files (for example, moving image files), and the burden on the server is small, so a larger amount of data can be widely used.

 [0091] As described above, in the present invention, when converting an audio information file into text, it may be divided by a predetermined time unit! / (See FIG. 2 etc.) or may be divided by a predetermined number of characters! / (See Fig. 7 etc.) and may be separated by the number of specified phrases or words (see Fig. 8 etc.). These are appropriately determined in consideration of the processing capability when converting the voice information file to text and the processing capability such as whether the text-converted file can be recognized as a character, a phrase or a word, etc. Select with power S.

 [0092] Thus, in the present invention, high-accuracy voice information retrieval with a high search speed and a low burden on the server is achieved by registering a voice information file as text data in a certain unit in the database. Can be done. Therefore, the present invention can be used widely and easily for various databases.

 FIG. 11 is a conceptual diagram for explaining an example of the speech information retrieval system according to the present invention.

The speech information retrieval system according to the present invention includes a text conversion server 8, a database server 9, a web server 10, a speech information file storage server 11, and a management server 12. The system connected by the network 13.

 The text conversion server 8 is a server that converts a voice information file into text, creates a content information file, and registers it in a database. The database server 9 is a server that stores the text-converted information file 91 and the content information file 92. The Web server 10 is a server that is connected to the user's terminal 101 and exhibits a search condition input from the user's terminal 101 and a function for displaying and outputting the search result. The audio information file storage server 11 is a server that stores an audio information file (including a video information file) 111. The management server 12 is a server that manages system failures such as each server and the network 13.

 [0096] Also in such a system, the above-described accuracy improvement mode can be operated as required. If the accuracy improvement mode is to function, set the accuracy improvement mode to “valid” (see Figure 6). As a result, playback is performed from the time stamp of the record to be searched, and the location that matches the keyword is specified. Then, the sum of the time until matching and the time of the time stamp is set as the playback start position for the user.

 [0097] If the accuracy improvement mode is not desired to function, the accuracy improvement mode is set to "invalid". As a result, playback starts from the time stamp of the record to be searched (that is, from the beginning of the record to be searched).

 In the present invention, the database capability can be adjusted by shortening the time required for search or improving the search accuracy by adjusting settings as appropriate. For example, when the accuracy improvement mode is set to be effective, the search accuracy of voice information is improved, and a position where a user's desired keyword is output (spoken) can be detected, and a more accurate reproduction start position can be determined. Searchable. On the other hand, by disabling the accuracy improvement mode, the time required for the search can be shortened because the step of the accuracy improvement mode is unnecessary, and the load on the server of the search system can be reduced. That is, according to the present invention, it is possible to appropriately construct a search system in consideration of the type and amount of information file to be searched, a certain hardware environment, and the hardware environment to be used.

Furthermore, the information recorded in the database used in the present invention is not limited to the audio information file (or moving image information file), and for example, a text information file is also recorded. You can also In the present invention, since voice information is converted into text and registered in the database, not only the voice information file but also the text information file can be simultaneously searched. That is, in the present invention, by registering not only the audio information file but also the text information file, the audio information file (or moving picture information file) and the text information file can be simultaneously stored by executing one search. You can search.

 [0100] Further, in the present invention, a step of designating a search target before the search execution may be separately provided. As a result, it is possible to make an appropriate selection in consideration of the user's purpose of use, usage environment, etc., such as a search for only audio information files or a search for only text information files.

 [0101] In the present invention, a function may be provided in which a plurality of contents can be searched simultaneously when performing a search process. By performing simultaneous search for multiple contents, the time required for search processing can be reduced. In this case, it is possible to appropriately determine whether to provide this function in consideration of the processing capacity of the computer or the like used in the present invention. Industrial applicability

 [0102] The voice information search program according to the present invention can be used for management, tabulation, and search of various voice information files. In particular, a multimedia-related database for managing a voice information file with a large amount of information, It can be used by incorporating a large number of information files into a search engine on the Internet that is the search target.

 Brief Description of Drawings

 FIG. 1 is a conceptual diagram for explaining a speech information retrieval program according to the present invention.

 FIG. 2 is a flowchart for registering a database in the first embodiment of the audio information program according to the present invention.

 FIG. 3 is a conceptual diagram illustrating a basic information table of audio information files registered in a database in the first embodiment.

 FIG. 4 is a conceptual diagram illustrating content data of a voice information file registered in a database in the first embodiment.

FIG. 5 is a flowchart when a search process is performed for a database in the first embodiment. FIG. 6 is a flowchart for outputting a detection result in the first embodiment.

 FIG. 7 is a flowchart for registering a database in the second embodiment of the audio information program according to the present invention.

 FIG. 8 is a flowchart for registering a database in the third embodiment of the audio information program according to the present invention.

 FIG. 9 is a flowchart when a search process is performed for a database in the third embodiment.

 FIG. 10 is a flowchart for outputting a detection result in the third embodiment.

FIG. 11 is a conceptual diagram for explaining an example of a speech information retrieval system according to the present invention. Explanation of symbols

 1 Text conversion function for audio information files

 2 Audio information file playback time measurement function

 3 Database registration function

 4 Search condition determination function

 5 Search processing function

 6 Search result output function

 7 Database

 8 Text conversion server

 9 Database Sano

 10 Web server

 11 Audio information file storage server

 12 Management server

 13 network

Claims

Claims [1] The step of converting the voice element recorded in the voice information into a character string, the step of measuring the playback time of the voice information, and the playback time of the voice information being longer than a predetermined time, The character 歹 IJ of the voice information is registered in the database by dividing it every predetermined time, and if the reproduction time of the voice information is less than the predetermined time, it is registered in the database without dividing the character string of the voice information. Determining a search item for searching the database and a search condition for the search item; executing a search for the database according to the search condition; and A voice information retrieval program that causes at least a computer to execute a result output step; [2] converting a voice element recorded in voice information into a character string; measuring a number of characters in the voice information; and if the number of characters in the character string is greater than a predetermined number of characters, the character string is If the number of characters in the character string is equal to or less than the predetermined number of characters, register the characters in the database without dividing the character string. A search item for performing the search, a search condition for the search item, executing a search for the database according to the search condition, and outputting a result of the search at least A voice information retrieval program to be executed by a computer. [3] The audio information search program according to claim 1 or 2, wherein the computer is caused to execute at least the following steps in order to execute a search on the database.  (1) A step of combining two pieces of audio information before and after the divided audio information, (2) The character specified by the search condition in the combined two voice information character strings Determining whether it is included,  (3) A step of determining whether the character string specified by the search condition is included in the character string of each voice information in the character strings of the two separated voice information. [4] The speech information search program according to any one of claims 1 to 3, wherein the computer executes at least the following steps in order to output the search result. .  (a) The step of designating the recording time position of the voice information corresponding to the character specified in the search condition is recorded! /  (b) a step of designating a recording time position retroactive by a predetermined time before the designated recording time position as a reproduction start position;  (c) The step of reproducing the designated reproduction starting position force voice.  [5] A computer-readable recording medium on which the speech information retrieval program according to any one of claims 1 to 4 is recorded. [6] means for converting a voice element recorded in voice information into a character string;  Means for measuring the playback time of the audio information;  If the playback time of the voice information is longer than the predetermined time, register the character 歹 IJ of the voice information in a database divided every predetermined time, and if the playback time of the voice information is less than the predetermined time, Means for registering in the database without dividing the character string of the voice information;  A search item for searching the database, a search condition determining means for the search item,  Means for executing a search on the database in accordance with the search condition; and means for outputting the search result;  Voice information retrieval system comprising at least [7] means for converting a voice element recorded in voice information into a character string;  A means for measuring the number of characters of the voice information; If the number of characters in the character string is larger than the predetermined number of characters, the characters constituting the character string are divided into predetermined characters and registered in the database, and the number of characters in the character string is determined. If the number of characters is less than or equal to the number of characters, means for registering the character string in the database without dividing  Means for determining a search item for searching the database and a search condition for the search item;  Means for performing a search on the database in accordance with the search condition; means for outputting a result of the search;  Voice information retrieval system comprising at least [8] A procedure for converting a voice element recorded in voice information into a character string,  Measuring the playback time of the audio information;  If the playback time of the voice information is longer than the predetermined time, register the character 歹 IJ of the voice information in a database divided every predetermined time, and if the playback time of the voice information is less than the predetermined time, Registering the voice information in the database without dividing the character string;  A procedure for determining a search item for searching the database and a search condition for the search item;  A procedure for performing a search on the database according to the search condition; a procedure for outputting a result of the search;  Voice information retrieval method comprising at least [9] A procedure for converting a voice element recorded in voice information into a character string;  A procedure for measuring the number of characters of the voice information;  When the number of characters in the character string is larger than the predetermined number of characters, the characters constituting the character string are registered in a database by dividing the character string by the predetermined number of characters, and when the number of characters in the character string is less than the predetermined number of characters , A procedure for registering in the database without separating the character string,  A procedure for determining a search item for searching the database and a search condition for the search item; A procedure for performing a search on the database according to the search condition; a procedure for outputting a result of the search; Voice information retrieval method comprising at least