JP4851081B2 - Information retrieval device - Google Patents

Description

  The present invention relates to an information search apparatus suitable for use in, for example, searching for personal data of company employees or the like by first and last name in a call center (telephone reception) of a device manufacturer. Specifically, for example, when a person's first and last names called up by an operator are recognized by voice and the person data is automatically searched, the display at the time of the search is performed satisfactorily.

  In an information search apparatus using a conventional speech recognition method, for example, a word that is a keyword in a conversation is searched and data related to the word is displayed (for example, refer to Patent Document 1). .

That is, in the invention of Patent Document 1, the presence or absence of a predetermined keyword in the utterance of the operator is determined, and information retrieval is performed when the keyword is detected.
JP-A-8-248987

  For example, in a call center (telephone reception) of an equipment manufacturer, there are cases where an inquiry from a customer is received and an inquiry from an in-house sales employee is received. Such inquiries from employees will be provided with more detailed information than inquiries from general customers, but the information includes information that should not be communicated to general customers. there is a possibility. Therefore, it is necessary to confirm whether or not the inquirer is a company employee.

  In this case, as a conventional means, for example, a method of inputting a surname given by the other party using a keyboard and searching a database storing employee personal data by the input is adopted. However, with such a means, it is not easy to operate such a keyboard while continuing to answer a telephone, for example, and skill is required to do this. In addition, a situation in which a smooth search cannot be performed due to a mistake in hearing occurs.

  On the other hand, in the above-mentioned Patent Document 1, for example, when the keyword is uttered using the name, performance, model number, etc. of the device as a keyword, information on the device is retrieved by voice recognition. ing. According to this, skill such as keyboard operation or the like is not necessary, and keywords that are uttered when the operator repeats are expected to be accurately searched.

  However, the above-described invention of Patent Document 1 detects a keyword specified in advance and performs a search, and is effective for searching for a keyword in which a device name, performance, model number, and the like are specified. However, for example, when searching for person data by voice recognition of first and last names, there are many names with similar names and similar names such as “Kato” and “Kano”. Judgment proved extremely difficult.

  For example, when searching for person data by voice recognition from a database, the reading information of first and last names originally registered in the database is the target of voice recognition, but often the reading of first and last names is not registered. is there. In that case, for example, if registration of Roman letters is essential, the reading will be guessed from the Roman letters, but there are various fluctuations in the Roman letter notation, and it can always be converted into correct reading information It was not limited.

  That is, for example, “Sato” has notations such as “sato”, “satou”, and “satoh”. If you read the Roman letters straight, The remaining two are converted to “Sato” or “Sato”. “Koichi” can be “koichi”, “kouichi”, “kohichi”, and the like. In this case, “koichi” is converted to “koichi” and “kohichi” is converted to “kohichi”. Furthermore, in the case of “Junichi”, it is usually described as “junichi”, but if this is converted into a reading, it becomes “Junichi”.

  Such an example can only be mentioned, and in the end, it can be said that this is an essential problem when reading information is created mechanically only from Roman characters without using Chinese character information. The present invention has been made in view of the above points, and an object of the present invention is to provide an extremely good information retrieval apparatus used when, for example, human data is retrieved by voice recognition. It is.

In order to solve the above-described problems and achieve the object of the present invention, each of the inventions described in claim 1 includes at least one of a name, a furigana of the name, and a romanization of the name, and a predetermined For each of the plurality of search targets and a database having a plurality of search targets corresponding to information, and the search target does not include the furigana, from the Roman alphabet notation included in the search target, Reading information guessing means that guesses and creates a furigana, the furigana included in each of the plurality of search objects, and the degree of coincidence between the input user's voice and the recognized voice Speech recognition means for specifying a plurality of phonetics as candidates for phonetics, and each of the plurality of phonetics specified by the speech recognition means In the case where the search object including the furigana created by the reading information estimation means exists in at least one search object including the furigana, the at least one search object including the furigana includes A recognition result display means for displaying that there is the search object including the furigana created by the reading information estimating means, and one selected from the plurality of furiganas specified by the voice recognition means. An information search apparatus comprising: information display means for extracting the search target including the furigana from the database and displaying the predetermined information included in the search target.

In the information search apparatus of the present invention according to claim 2, the recognition result display means includes at least one name that is identified by the phonetic for each of the plurality of phonetics specified by the voice recognition means. When all of the two search targets include the phonetic created by the reading information estimation means, and only a part of the at least one search target including the name identified by the phonetic is the reading The display method is different depending on the case of including the furigana created by the information estimation means .

In the information search device of the present invention according to claim 3, the recognition result display means includes at least one of the search objects including the name identified by the phonetic specified by the voice recognition means. In order to display that the search target including the phonetic name created by the reading information estimation means is present, a display is performed with a specific symbol or icon attached before or after the character string representing the specified phonetic name. It is characterized by doing.

In the information search device of the present invention according to claim 4, the recognition result display means includes at least one of the search objects including the name identified by the phonetic specified by the voice recognition means. In order to display that the search target including the furigana created by the reading information estimation means is present, the font shape of the character string representing the identified furigana is changed and displayed. It is.

In the information search device of the present invention according to claim 5, the recognition result display means includes at least one of the search objects including the name identified by the phonetic specified by the voice recognition means. In order to display that the search target including the furigana created by the reading information estimating means is present, the font color of the character string representing the identified furigana is changed and displayed. It is.

In the information search device of the present invention described in claim 6, the recognition result display means includes at least one of the search objects including the name identified by the phonetic specified by the voice recognition means. In order to display that the search target including the furigana created by the reading information estimation means is present, the character string representing the identified furigana is displayed with a different background color. It is.

  According to the present invention described above, the reading method is estimated from the Roman alphabet information, and the estimated reading information is assigned to at least one of the entries in the database corresponding to each recognition result. The fact that it is displayed together with the recognition result informs the user that there is a possibility that it contains a conversion error, and is suitable information when searching for a correct answer from a plurality of recognition candidates. A search device can be provided.

  By the way, with respect to the above-mentioned fluctuations in the Roman alphabet, the fluctuations as described above are often quite similar to the correct answer in pronunciation, so even if the reading information was created mechanically from only the Roman letters, If the recognition target vocabulary is determined by combining the correct reading information, and voice recognition is performed based on the vocabulary, then typing “sato” will not only pronounce “sato” but also sound like “sato”. Can also be detected.

  Although this is a result because speech recognition is incomplete, in the case of the present invention, it is a merit, and so-called fuzzy search is realized as a result. In other words, this indicates that even if the reading information includes some errors, such candidates can be picked up by voice recognition.

  However, for example, if you want information about Koichi Sato and enter “Sato Koichi”, even if “Satoko Hichi” is displayed in the recognition candidates, the user does not know There is a risk that you may not realize that it represents a candidate. As a result, it is determined that there is no correct answer among the candidates, and if the input is performed again or the input is performed again, the same result may be obtained, which may eventually give up.

  The present invention has been made paying attention to such points. For example, by telling the user that there may be a conversion error, the user searches for a correct answer from among the recognition candidates. , So that they can take appropriate care. The present invention will be described below with reference to the drawings. First, FIG. 1 is a block diagram showing the overall configuration of an embodiment of a person data search system to which an information search apparatus according to the present invention is applied.

  In FIG. 1, a talker A who is an operator (input person) and a talker B who is a caller are having a conversation through a telephone line 200 via voice input / output units 100A and 100B which are telephones, for example. . On the other hand, the information retrieval apparatus 1 according to the present invention is constituted by a so-called personal computer. The information retrieval apparatus 1 is connected to a microphone (Mic) 2 that collects the voice of the operator (input person). Accordingly, the microphone 2 picks up only the voice of the speaker A who is an operator (input person).

  In addition, the information search apparatus 1 includes a waveform extraction unit 11 that extracts a waveform of the sound collected by the microphone 2 described above, a speech recognition unit 12 that recognizes the collected sound, and the recognized information. A search means 13 for searching the list based on the display 14, a display 14, and an input means 15 such as a keyboard and a mouse are provided. The waveform extracting means 11 to the input means 15 are connected to each other through an internal bus 16, exchange data with each other, and control the operation by a control device (CPU) 17.

  Furthermore, an employee database 4 is connected to the search means 13 of the information search apparatus 1 through, for example, an in-house LAN (Local Area Network) 3, and by accessing the employee database 4, employee personal data can be obtained. Therefore, as a preparatory stage, the search means 13 of the information search apparatus 1 is used to search the employee database 4 for person data of all employees to be searched, and the data set is stored in the information search apparatus 1 through the internal bus 16. It is stored in the hard disk 18 or the like.

  In such a person data search system, FIG. 2 further shows functional blocks of the information search apparatus 1 of the present invention centering on a control device (CPU) 17. In the following description, an embodiment in which claim 1, claim 2, and claim 3 are combined will be described. For example, a voice input of an employee's name + honorific name (san) is accepted, and an employee of that name (same name and same name) An information search system in which incidental information (employee number, affiliation, telephone number, fax number, etc.) of all employees is extracted from a database and displayed will be described.

  In FIG. 2, reference numeral 21 denotes an employee information database, which has an entry as shown in FIG. 3, for example. Reference numeral 22 denotes reading information estimation means, which automatically creates reading information by using Roman character information for the entry in the database for which the phonetic information is not registered. Further, 23 is a grammatical rule for speech recognition, which is described by, for example, a phoneme finite state automaton. Reference numeral 24 denotes an acoustic model for speech recognition, for example, an HMM corresponding to an unspecified speaker.

  Furthermore, 25 is a voice recognition means, which is a voice recognition device capable of calculating n candidates in descending order of recognition score with respect to the input voice. A recognition result display means 26 displays n recognition candidates (Hiragana character strings) obtained by the speech recognition means on a display device such as a display. Reference numeral 27 denotes information display means for extracting an entry corresponding to one candidate in the recognition result from the employee information database and displaying the detailed information of the employee on a display device such as a display.

  By the way, in the example of the database shown in FIG. 3 described above, for example, “Kato Sato” has a furigana defined. Therefore, in the case of “Kouichi Sato”, even if the Roman alphabet is “sato kouichi” Searched by “Koichi Sato”. On the other hand, “Kouichi Sato” and “Kouichi Sato” are not registered for furigana. For this reason, reading information is created based on the Roman alphabet. Here, “Sato Junichi” is created as “Satoh Koichi” because the Roman display is “satoh kouichi”, while “Sato Koichi” is created as “Sato Koichi”.

  Also, since furigana is defined in “Sato Kokichi”, it will be searched for “Sato Kokichi”, but “Sato Kokichi” has not been registered for furigana, and here the Roman letter is “sato koukichi”. Therefore, the reading information is created as “Sato Kokichi”. Then, speech recognition is performed to return the reading information of the names excluding the honorifics as the recognition vocabulary with the readings of the names (furigana) of all the databases added to each word as the recognition target vocabulary. Is called.

  So, for example, when searching for “Koichi Sato” and entering “Koichi Sato”, “Sato Koichi” with the correct voice recognition is selected as the first candidate, and the following is the second from the pronunciation It is assumed that “Sato Koichi” is selected as a candidate, “Koko Sato” is selected as the third candidate, “Sato Koichi” is selected as the fourth candidate, and “Sato Koichi” is selected as the fifth candidate. These candidates are selected from vocabulary registered as reading information in the database.

  In this case, there are three employees on the database who are given the first candidate “Satou Koichi” reading information, and the reading information (furigana) of the two of them is automatically created from the Roman alphabet. It has been done. The following candidates are one each on the database, and the third candidate “Sato Kokichi” has registered reading information (furigana), but the other second candidate “Sato” “Koichi”, the fourth candidate “Satokohichi”, and the fifth candidate “Satokohichi” are automatically created from Roman letters.

  Therefore, by displaying the presence / absence of reading information (phonetics) automatically created from these romaji notations, it is possible to inform the user that there may be a conversion error, for example.

  That is, for example, a candidate “contains reading information generated from Roman characters for some entries” such as “Satou Koichi” is displayed with a symbol “☆” indicating that. Also, no symbol is displayed for a candidate “reading information is registered for all entries”, such as “Sato Kokichi”. Furthermore, the symbol “★” indicating that is added to the candidate “the reading information is generated from Roman characters for all entries” such as “Sato Koichi”, “Sato Koichi”, “Sato Koichi”. To display.

  Therefore, in the information search apparatus 1 according to the present invention, when searching for person data, an image as shown in FIG. 4 is displayed on the display 14 of FIG.

  In FIG. 4, first, a “noise measurement” and “end” operation area 32 is provided on the left side below the title area 31 at the top, and the cursor is moved to these operation areas 32 to perform a click. Each processing operation is performed. Below this, a selection area 33 for voice input is provided. For example, there are options of “male and female common”, “male only”, “female only”, and “automatic selection”. Move the cursor to these options and click Each selection is performed by performing.

  Further below this, an area 34 for displaying the waveform of the sound extracted by the waveform extracting means 11 is provided. That is, in this area 34, the waveform of the sound collected by the microphone 2 is displayed. Therefore, the operator (input person) can confirm the volume of the sound input to the microphone 2 by looking at the waveform displayed in the area 34. Thereby, the operator (input person) realizes more appropriate voice input by adjusting the volume of his / her voice inputted to the microphone 2 and searches for person data by smoother voice recognition. Will be able to.

  As a result, sound is collected by the microphone 2 in FIG. 1, and the recognized sound is recognized by the sound recognition means 12. Here, according to the ambiguity of voice recognition, a plurality of candidates are raised from the one with the high degree of matching. Then, based on the result of the voice recognition, the search means 13 searches for the reading (pronunciation) of the first and last names from the list stored in the hard disk 18. Further, this search result is displayed in the right area 35 below the title area 31 of the image shown in FIG. In this area 35, for example, hiragana readings of a plurality of candidates are displayed in descending order of the matching degree, and a selection check column is provided at the left end of each hiragana display column.

  If the check box is further checked, the retrieved person data is displayed in the area 36 below the image. In the case of voice recognition of first and last names, there may be a plurality of person data with the same first and last names, but these data are displayed in parallel in the area 36. In addition, since speech recognition is performed by reading first and last names, items with different kanji notations are also displayed in parallel. Furthermore, when the number of retrieved person data exceeds the number that can be displayed in the area 36, the scroll bar 37 can display the amount exceeding the number that can be displayed.

  In the display area 38, the number (number of people) of the retrieved person data is displayed. Furthermore, the display area 39 displays “standby”, “starting end detection”, “end end detection”, “recognition”, and the like, and displays the operation state of each device.

  In other words, in the information search apparatus 1 of the present invention, when the operator (input person) speaks and the voice collected by the microphone 2 is recognized by the voice recognition means 12, the voice recognition means 12 shows the degree of coincidence of voice recognition. Hiragana of candidate readings are listed in order from the highest. Then, the operator (input person) puts a check in the check box of the desired reading from the hiragana of these readings, and displays the person data retrieved by the reading in the area 36. Thereby, the operator (input person) can confirm the person data by looking at the display of the area 36.

  Further, in the information search apparatus 1 of the present invention, in the above-described area 35, for example, as described above, “for some entries, the reading information includes those generated from Roman characters” A symbol “☆” indicating that is displayed together, and a symbol “★” indicating that is displayed together with a candidate “reading information generated from Roman characters for all entries”. .

  That is, in FIG. 4, for example, the first candidate “Satou Koichi” “includes reading information generated from Roman characters for some entries”. Are displayed together. In addition, since the third candidate “Sato Kokichi” “reading information is registered for all entries”, no symbol is displayed. Still other candidates “Sato Koichi”, “Sato Koichi”, and “Sato Koichi” have “the reading information is generated from Roman characters for all entries”, so the symbol “★” representing that fact Are displayed together.

  In FIG. 4, since the check box for the first candidate “Kouichi Sato” is checked, the person data searched for “Kouichi Sato” in the area 36 is displayed in parallel. Yes. In the three person data, a symbol “★” indicating that “the reading information is generated from Roman characters” is also displayed for the second and third person data.

  In this way, by displaying the presence / absence of reading information (phonetics) automatically created from Roman notation, it is possible to inform the user that there may be a conversion error, for example.

  In this case, the symbol “☆” is displayed on the first candidate “Sato Koichi”, which means that “for some entries, the reading information includes those generated from Roman characters”. This means that even if "reading information is generated from Roman characters", the reading itself is possible. On the other hand, the symbol “★” is displayed in the second candidate “Sato Koichi”, which is “for all entries, reading information is generated from Roman characters”. Therefore, the reading information itself may be incorrect.

  On the other hand, since no symbol is displayed in the third candidate “Sato Kokichi”, “reading information is registered for all entries”. Therefore, in this case, since the reading information is considered to be correct, the third candidate whose reading information is different from the desired “Satou Koichi” can be excluded from the search target performed by the user. In other words, in this case, there is a possibility for the first candidate “Sato Koichi”, the second candidate “Sato Koichi”, the fourth candidate “Sato Koichi”, and the fifth candidate “Sato Koichi”. Is left behind.

  Therefore, for example, when a check is put in the check box of “Satoko Hichi” of the fifth candidate, as shown in FIG. 5, the person data of “Kouichi Sato” is displayed in the area 36, and the desired person data is displayed. You can search.

  That is, in the above-described embodiment, if only a part of the entries in the database corresponding to the hiragana character string obtained as a recognition result is unregistered (that is, the remaining characters are correctly input) ) Means that there is an actual employee who has the same reading as the reading information automatically converted from Roman characters (same name and same name), and at least it is not appropriate as a personal name.

  On the other hand, when all the entries in the database corresponding to the hiragana character string obtained as a recognition result are unregistered phonetics (that is, all entries use reading information automatically generated from Roman characters) ), The character string may have been read inappropriately as a person's name, and it is highly likely that it contains errors compared to the former.

  Therefore, by changing the result display method for both of them, it is possible to inform the user of the possibility of an error in the hiragana character string.

  Accordingly, in order to solve such a problem, in the first aspect of the present invention, when the reading information is automatically generated from Roman characters, that is, that is, “displayed as a recognition result”. Some hiragana strings are automatically converted from Romaji, so they may be wrong. " If it is reported that there is a possibility of including the conversion error as described above, it should be possible to take appropriate consideration even when the user searches for a correct answer from among the recognition candidates.

  Further, in claim 2 of the present invention, mention is made of a device for displaying the possibility that the hiragana character string includes an error in two stages. Furthermore, in claim 3 of the present invention, mention is made of a device for adding and displaying some symbols, icons, etc. to the hiragana character string of the recognition result.

  FIG. 6 shows an embodiment in which the font shape is changed (the font shape is italic) as disclosed in claim 4 instead of claim 3. Further, as disclosed in claims 5 and 6, an embodiment in which the font color and the background color are both changed (the font color is changed from black to white and the background color is changed from white to black) is shown. 7 shows.

  In this way, as a specific method of displaying that there is a possibility of including an error, the display is performed by adding some symbols or icons to the hiragana character string of the recognition result, changing the font shape, Various methods such as changing the font color and background color can be considered. In these methods, a candidate that is likely to contain an error is explicitly displayed.

  Thus, according to the information retrieval apparatus of the present invention, the reading method is estimated from the Roman character information, and the estimated reading information is assigned to at least one of the entries in the database corresponding to each recognition result. In this case, the fact that it is displayed together with the recognition result informs the user that there may be a conversion error, and is suitable for searching for a correct answer from a plurality of recognition candidates. Can be provided.

  The information search apparatus of the present invention is not limited to the employee search at the company representative reception (telephone exchange) described above, but stores customer searches for mail order companies, product names, item numbers, sales prices, and the like. This is a database that can be used effectively for work that teaches item numbers, sales prices, etc. from product names by phone.

  The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention.

It is a block diagram which shows the structure of one Embodiment of the person data search system to which the information search device by this invention is applied. It is a functional block diagram of one embodiment of an information retrieval device according to the present invention. It is a block diagram of the database for the description. It is a block diagram of the display screen for the description. It is a block diagram of the display screen for the description. It is a block diagram of the display screen for the description. It is a block diagram of the display screen for the description.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Voice recognition system, 2 ... Microphone, 3 ... LAN (Local Area Network), 4 ... Employee database, 11 ... Waveform extraction means, 12 ... Voice recognition means, 13 ... Search means, 14 ... Display, 15 ... Keyboard, 16 DESCRIPTION OF SYMBOLS ... Internal bus, 17 ... Control device, 18 ... Hard disk, 21 ... Employee information database, 22 ... Reading information estimation means, 23 ... Grammar rules for voice recognition, 24 ... Acoustic model for voice recognition, 25 ... Voice recognition means , 26 ... recognition result display means, 27 ... information display means

Claims (6)

A database having a plurality of search targets each corresponding to predetermined information and at least one of a name, a furigana of the name, and a Roman notation of the name;
For each of the plurality of search targets, when the phonetic is not included in the search target, reading information estimation means that guesses and creates the phonetic name from the Roman alphabet notation included in the search target;
A voice that identifies a plurality of phonetics as candidates of the phonetic recognized by the input voice based on the degree of coincidence between the phonetic included in each of the plurality of search objects and the input user's voice Recognition means;
For each of the plurality of phonetics specified by the speech recognition means, the search target including the phonetic created by the reading information estimation means exists in at least one of the search targets including the furigana. Is a recognition result display means for displaying that the search object including the phonetic name created by the reading information estimation means is present in at least one of the search objects including the furigana;
Information display means for extracting, from the database, the search target including the one phonetic selected from the plurality of phonetics specified by the voice recognition means, and displaying the predetermined information included in the search target. And comprising
An information retrieval apparatus characterized by that. The information search device according to claim 1,
The recognition result display means includes, for each of the plurality of phonetics specified by the voice recognition means, all of at least one of the search objects including the phonetic includes the phonetic created by the reading information estimation means. A display method is different between a case and a case where a part of at least one of the search objects including the furigana includes the furigana created by the reading information estimation means.
An information retrieval apparatus characterized by that. The information search device according to claim 1 or claim 2, wherein
The recognition result display means indicates that the search target including the phonetic created by the reading information estimation means exists in at least one search target including the phonetic specified by the voice recognition means. In order to display, a specific symbol or icon is accompanied or displayed before or after the character string representing the specified phonetic character,
An information retrieval apparatus characterized by that. The information search device according to claim 1 or claim 2, wherein
The recognition result display means indicates that the search target including the phonetic created by the reading information estimation means exists in at least one search target including the phonetic specified by the voice recognition means. In order to display, the font shape of the character string representing the specified furigana is changed and displayed.
An information retrieval apparatus characterized by that. The information search device according to claim 1 or claim 2, wherein
The recognition result display means indicates that the search target including the phonetic created by the reading information estimation means exists in at least one search target including the phonetic specified by the voice recognition means. In order to display, the font color of the character string representing the specified furigana is changed and displayed.
An information retrieval apparatus characterized by that. The information search device according to claim 1 or claim 2, wherein
The recognition result display means indicates that the search target including the phonetic created by the reading information estimation means exists in at least one search target including the phonetic specified by the voice recognition means. In order to display, the background color of the character string representing the specified furigana is changed and displayed.
An information retrieval apparatus characterized by that.

Images