JP2003228394A - Noun specifying device using voice input and method thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a noun specifying device which can automate large-scale specified object operation that a human operator has handled before such as the construction of a call center using voice recognition and realize higher- precision and higher-efficiency speedy retrieval from a database consisting of many objects not as a substitute for the human operator, and uses voice input. <P>SOLUTION: The noun specifying device which performs voice recognition processing for contents that a user inputs in voice by using a voice recognizing device and narrows down results of the voice recognition processing is characterized in that it uses information based upon differences in notation as additional information for the narrowing-down processing and specifies a noun that the user inputs in voice among many retrieval objects. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a search for uniquely specifying information desired by a user from a very large number of similar search targets and a narrowing down to the specification.

[0002]

2. Description of the Related Art Conventionally, when information that a user wants from a large number of search targets is uniquely specified by using additional information, the attributes of the information are used to classify the information in advance. , Ask the user for the attribute value of the attribute that the target information has, narrow down the target information based on the attribute value answered by the user, and by this, select the target information from many candidates. The method of narrowing down and uniquely identifying is usually
Has been adopted.

Generally, when the information to be searched has attributes such as a prefecture name, a city / ward county name, a town / village name like an address in Japan, and the attributes have a hierarchical structure. ,
If you have attributes such as department, year of employment, gender, etc. like an employee list, ask the user for the attribute value for each attribute, narrow down the search candidates based on the attribute value answered by the user,
It is possible to specify.

[0004]

However, when the search target is information having no attribute, and when there are many candidates and they are very similar to each other, the search candidates are narrowed down and specified. There is no additional information available to do this.

In particular, when the user inputs a noun by voice,
Since there is no guarantee that the accuracy of the voice recognition device will always be 100%, there is no way to resolve the ambiguity, and the candidates returned by the recognition device are presented to the user in order, and the user obtains the correct answer. Presentation to the user is repeated until it is asked. Repeating this presentation is said to be very stressful to the user. Therefore, there is a problem that the information searched by the current search system using voice recognition is very limited, and the number of the searched objects is also very small.

The present invention makes it possible to automate a large-scale target specific task, which was conventionally handled by a human operator, such as the construction of a call center using voice recognition, and is not only a substitute for the human operator. An object of the present invention is to provide a noun identification device that utilizes voice input, which can realize a more accurate, more efficient, and quick search when searching a database with a large number of objects. .

[0007]

The present invention utilizes Japanese notation as additional information for narrowing down a very large number of information sets.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION First, let us consider a case in which the first name and last name of a voice-input Japanese individual are to be searched.

Japanese personal surnames and personal names are intended for 40 million fixed telephone subscribers, and if the readings are uniquely counted, there are about 180,000 different personal surnames.
It is known that there are about 160,000 types of personal names.

Generally, a voice recognition device is used to recognize with a range of accuracy and processing speed that the user can accept.
The number of identifiable targets is about several hundreds to 1,000. Under such circumstances, it is said that the objects of more than 100,000 kinds are very large in scale, and it is difficult to accurately recognize them by using only the voice recognition device.

On the other hand, the voice input of the personal first and last name is an indispensable target in the customer identification in the customer care service such as call center business, and if the voice input can be automatically specified. , There is a great demand.

However, the personal surname and personal name do not have hierarchical attributes.

In the embodiment of the present invention, when paying attention to the notation such as the character string constituting the individual surname, the number of characters, etc., the reading of the leading kanji is particularly noted, and the reading information of the leading kanji is added as follows. Consider using it as information.

There are about 300 types of leading kanji used for 180,000 personal surnames. Therefore, if information indicating what kind of kanji is the leading kanji is acquired as additional information for narrowing down, it is the target (search target) that should be voice-recognized.
Can be reduced to 1/300 of the speech recognition target in the conventional method.

Therefore, voice recognition processing is performed on the user's input, and the result is output. As a result of this recognition processing, a candidate that is output with a very high reliability (likelihood) is likely to be the correct answer, so it is presented to the user and confirmed.

When a plurality of highly reliable candidates are output,
Alternatively, if no candidate is output, the user is queried as the additional information for narrowing down the phonetic kana of the first kanji in the noun that is input by voice.

When the reading of the first kanji in the noun that is input by voice is acquired, the individual surname having the reading of that kanji is
Extracted from the search target database, perform voice recognition processing again only for the extracted search target, or when multiple candidates are output, the first Kanji from the multiple candidates has the corresponding reading. If a search candidate exists, it is presented to the user and confirmed. By doing so, the voice recognition result can be narrowed down not only by using only the voice recognition technology but also by using the additional information, thereby improving the reliability of the recognition result.

Next, the above-described embodiment will be described by taking as an example the case where a specific individual surname "Masao" is specified.

FIG. 1 is a flow chart showing the operation of the above embodiment.

First, when the search device (the noun determination device 1 shown in FIG. 7) requests the user to input the personal surname to be searched (S1), "masu" is voice-input (S).
2).

FIG. 2 is a diagram showing an example of the recognition target list L1 in the above embodiment.

The search device (noun determination device 1) holds in advance the personal surname list to be searched as the recognition target list L1. In this case, according to the voice recognition device to be used,
In other words, it is shaped and stored according to the input form.

At the same time, for each personal surname, the reading kana of the leading kanji and the character string information indicating the number of characters of the personal surname are held.

The retrieval device (noun determination device 1) performs a voice recognition process on the voice input "masu" (S).
3) From the recognition target list L1, correct answer candidates are output in the order of high reliability (highest likelihood), for example, "1: Matsuo, ........., 4: Matsuda" as shown below. Yes (S4).

1: Matsuo (Matsuo, tail), 2: Matsuno (Matsuno, Mino), 3: Masuo (Masuo, Masuo), 4: Matsuda (Matsuda, Matsuta). It is a figure which shows the example of the Chinese character description candidate list L2 in an example.

The kanji notation candidate list L2 includes voice-inputted search targets (sato, suzuki, tanaka, etc.), its kanji notation information (Sato, Sato, Sato, sugar, etc.), and its leading kanji reading information ( It is a list that corresponds to each other.

With respect to each recognition result, a kanji notation candidate list L2 that can be considered from the personal surname list is listed in parentheses.

[0028] In order to specify the intended family name of the user,
The search device (noun determination device 1) asks the user "What do you read the first kanji?" (S5).

The user answers the reading of the character "increase" such as "increase (increase)" or "elevate (increase)" (S6).

FIG. 4 is a diagram showing a kanji reading target list L3 prepared in advance in the search device (noun determination device 1) according to the above embodiment.

The kanji writing candidate list L2 is a list in which readings of kanji, information of kanji candidates having the readings, and reading information of the leading kanji are associated with each other.

After the user answers the reading of the character "masu" (S6), the retrieval device (noun determination device 1) uses the kanji reading target list L3 shown in FIG. Perform (S7).

FIG. 5 is a diagram showing an example of the correspondence table L4 between the leading kanji reading and the personal surname in the above embodiment.

Correspondence table between leading kanji reading and individual surname and L4
Is a list in which the reading of the leading kanji, the kanji corresponding to the reading of the leading kanji, and the personality of using the kanji correspond.

Utilizing the correspondence table L4 between the leading kanji reading and the personal surname, the kanji having the reading of the recognition result and the individual family name having the kanji at the head are listed. There is ambiguity in the recognition of "fluttering", and the result is 1: "fluffing" (increasing), 2: "hiring" (cooling), 3: "flying" (growing, shining, prospering) Suppose

FIG. 6 is an explanatory diagram showing the operation in the case of identifying the individual surname "Masuo" based on the voice input "Masuo" in the above embodiment.

Here, the kanji candidates that have the reading "fueru" in response to the voiced "fueru" are shown in FIG.
As shown in (3), all the results are obtained by searching the kanji reading target list (S8) and recognizing the first character of the personal surname,
The relationship between the two is checked together with the recognition result of the leading Kanji reading shown in FIG. 6 (4) (S9).

That is, the individual surname candidate having the kanji obtained as a result of recognizing the reading of the leading kanji is the output result, "Matsuo", "Matsuno", and "Masuo". , "Matsuda" is checked (S10). In this case, "masuo (masuo)" is shown in Figure 6.
As shown in (5), they are listed as applicable candidates.

That is, the above-described embodiment is an embodiment in which the Japanese notation is used as the additional information for narrowing down, regarding a very large number of information sets.

The above-mentioned "Japanese notation" means information such as the first character in the Hiragana notation, the second character from the beginning, the last character, etc., which is the specific character counted from the beginning. Information on the number of characters when written in hiragana, and the characteristics of the kanji used when writing kanji, for example,
Reading the first kanji, reading the second kanji, etc., reading information for the specific kanji from the beginning, or radical of the first kanji, etc.
It is used as additional information for narrowing down.

Then, the candidates of the checked result are displayed (S10), the user is asked a question (S11), the answer from the user is obtained (S12), and if the displayed candidate is correct, the displayed candidate is displayed. Confirm with.

On the other hand, if the displayed candidate is not correct and the user judges (S12), the second or the like the nth (n is an integer of 2 or more) kanji among the personal surnames input by voice. About how to read, ask the user to input voice, recognize the voice, search for Kanji candidates, and ask the user (S1
3).

FIG. 7 is a block diagram showing a noun determination device 1 which is an embodiment of the present invention.

The noun determination device 1 includes a voice input unit 2, a voice recognition unit 3, a voice recognition software 3S, a voice recognition result output unit 4, a dialogue control unit 6, a voice output unit 7, and a voice output. Software 7S and system database 8
Have and.

In the noun determination device 1, the voice of the user P input through the voice input unit 2 is sent to the voice recognition unit 3, and the voice recognition unit 3 performs a system for voice recognition processing of the input voice. The database 8 is used. Further, the voice recognition unit 3 executes the recognition process on the voice input by the user P by using the voice recognition software 3S.

The system database 8 includes a search database 81, a search auxiliary database 82, and YES / N.
And an O database 83.

The search database 81 is a database in which a plurality of individual family names are registered as search words.

The search assisting database 82 is a recognition target list L1 related to all individual surnames (search target words) registered in the search database 81 and a kanji writing candidate list L.
2. A database storing a kanji reading target list L3 and a correspondence table L4 for reading the first kanji and individual surnames.

The YES / NO database 83 contains the contents that the user P responded to (for example, YES / NO, YES / NO).
It is a database that recognizes (NO).

The voice recognition software 3S selects the search database 81 or the search auxiliary database 82 from the system database 8 in accordance with the scene of processing of the search device (noun determination device 1).

When a search word is input by voice, the search auxiliary database 82 is referred to, and when the response to the user P for the correctness confirmation is recognized, the YES / NO database 83 is referred to.

Further, the voice recognition section 3 uses the voice recognition software 3S in the voice recognition processing, and the voice output section 7 uses the voice output software 7S in the voice output.

That is, in the above embodiment, in the noun specifying device for performing the voice recognition processing of the content input by the user by the voice using the voice recognition device and narrowing down the result of the voice recognition processing, as the additional information for narrowing down, This is an example of a noun identification device that utilizes voice input to identify a noun that a user has voice input from a large number of search targets, using information based on the difference in notation.

Further, in the above-described embodiment, the contents input by the user by voice are subjected to the voice recognition processing by using the voice recognition device,
In a noun identification device that narrows down the result of voice recognition processing, an individual surname input requesting unit that requests a user to input a personal surname by voice, a voice recognition unit that recognizes an input voice by voice, and the input individual Based on at least one of the notation of the family name, the notation input requesting means for requesting the user to make a voice input, the voice input personal surname, and the voice input the notation. It is an example of a noun identification device using voice input having a personal surname characteristic means for identifying a personal surname input by voice.

In this case, the above-mentioned notation is information on the reading of the first or nth (n is an integer of 2 or more) kanji of the kanji used when the kanji is written, and is the radical of the first kanji. Further, it is the information of the beginning or the nth (n is an integer of 2 or more) or the end of the hiragana used when the hiragana is written, and the number of characters when the hiragana is written.

Further, in the process of narrowing down the target nouns by the user, a group is created in which only personal surnames that are used more than a predetermined number among personal surnames having a predetermined kanji as the leading kanji are used. Alternatively, only that group may be searched as the search target.

Furthermore, the above embodiment can be understood as a program invention, that is, in the above embodiment, the contents input by the user by voice are subjected to voice recognition processing using a voice recognition device, When narrowing down the results of recognition processing, an example of a program that causes the computer to execute the procedure to specify the noun spoken by the user from a large number of search targets, using information based on the difference in notation as additional information for narrowing down Is. Further, in the above-described embodiment, when the contents input by the user by voice are subjected to the voice recognition processing using the voice recognition device and the result of the voice recognition processing is narrowed down, the user is requested to input the personal surname by voice. At least one of a personal surname input request procedure, a voice recognition procedure for recognizing an input voice, and a notation of the input personal surname
The voice input personal name, based on the voice input personal name and the voice input personal name, the voice input input request procedure for requesting the user to input voice. 3 is an example of a program that causes a computer to execute the personal surname characteristic procedure.

The above program may be recorded in a predetermined recording medium. In this case, FD, CD,
DVD, HD, and semiconductor memory are examples of the predetermined recording medium.

[0059]

According to the present invention, there is an effect that it is possible to automate a large-scale target specific task which was conventionally handled by a human operator, such as construction of a call center utilizing voice recognition. Not only as a substitute for a human operator, but also when searching a database with a large number of targets, the notation feature is used as a search key, so that more accurate, more efficient, and faster searches can be realized. It has the effect of being able to.

[Brief description of drawings]

FIG. 1 is a flowchart showing the operation of an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a recognition target list L1 in the above embodiment.

FIG. 3 is a diagram showing an example of a Chinese character notation candidate list L2 in the above embodiment.

FIG. 4 is a diagram showing a kanji reading target list L3 prepared in advance in the search device (noun determination device 1) in the embodiment.

FIG. 5 is a diagram showing an example of a correspondence table L4 for reading leading kanji and individual surnames.

[Fig. 6] "masu" input by voice in the above embodiment
FIG. 10 is an explanatory diagram showing an operation in the case of identifying the individual surname “Masao” based on the.

FIG. 7 is a block diagram showing a noun determination device 1 according to the first embodiment of the present invention.

[Explanation of symbols]

1 ... Noun determination device, 2 ... Voice input section, 3 ... voice recognition unit, 3S ... software for voice recognition, 4 ... voice recognition result output unit, 6 ... Dialogue control unit, 7 ... voice output section, 7S ... software for voice output, 8 ... System database.

Claims (18)

[Claims] 1. A noun identification device that performs voice recognition processing of a voice input content by a user using a voice recognition device and narrows down the result of the voice recognition processing. A noun specifying device using voice input, characterized in that a noun input by a user by voice is specified from a large number of search targets using. 2. A noun identification device for performing voice recognition processing of contents input by a user using a voice recognition device using a voice recognition device and narrowing down the result of the voice recognition process, wherein the user is required to input the personal surname by voice. Requesting personal surname input requesting means; voice recognizing means for recognizing input voice; requesting the user to voice input for at least one of the notations of the personal name input by voice Notation input requesting means ;; personal surname characteristic means for specifying the voice input personal surname based on the voice input personal surname and the voice input personal notation; A noun identification device that uses voice input. 3. The claim according to claim 1 or 2, wherein the notation is information on the reading of the head or nth (n is an integer of 2 or more) of the kanji used when the kanji is written. A noun identification device that utilizes characteristic voice input. 4. The noun identification device according to claim 1 or 2, wherein the notation is a radical of the first Chinese character. 5. The method according to claim 1 or 2, wherein the notation is information at the beginning or nth (n is an integer of 2 or more) or information at the end of the hiragana used in the hiragana notation. A noun identification device that utilizes characteristic voice input. 6. The noun identifying apparatus utilizing voice input according to claim 1 or 2, wherein the notation is information on the number of characters in Hiragana notation. 7. The method according to claim 1 or 2, wherein, in the process of narrowing down the intended noun by the user, more than a predetermined number of personal surnames having a predetermined kanji as the leading kanji are used. A noun identification device using voice input, which is characterized by making a group of only individual surnames and searching only that group as a search target. 8. A noun identifying method for performing voice recognition processing on a content input by a user by using a voice recognition device and narrowing down the result of the voice recognition processing. A noun identification method using voice input, which is characterized by identifying a noun that the user has input by voice from a large number of search targets. 9. A noun identifying method for performing voice recognition processing of a voice input content by a user using a voice recognition device and narrowing down the result of the voice recognition processing, wherein the user is required to voice input an individual surname. Requesting personal last name input step; voice recognition step of recognizing input voice; requesting the user to voice input at least one of the notations of the personal last name input by voice A notation input request step; a personal surname characteristic step of specifying the voice input individual surname based on the voice input personal surname and the voice input the notation. A noun identification method that uses voice input. 10. The claim according to claim 8 or claim 9, wherein the notation is information on reading of the head or nth (n is an integer of 2 or more) of the kanji used when the kanji is written. A method for identifying nouns using a featured voice input. 11. The noun specifying method using voice input according to claim 8 or 9, wherein the notation is a radical of the leading kanji. 12. The method according to claim 8 or 9, wherein the notation is the first or nth (n is an integer of 2 or more) information or the last information of the hiragana used in the hiragana notation. A method for identifying nouns using a featured voice input. 13. The noun identifying method using voice input according to claim 8 or 9, wherein the notation is information on the number of characters in Hiragana notation. 14. The method according to claim 8 or 9, wherein, in the process of narrowing down the intended noun by the user, more than a predetermined number of personal surnames having a predetermined kanji as the leading kanji are used. A noun identification method using voice input, characterized by creating a group that collects only individual surnames and searching only that group as a search target. 15. When the user performs voice recognition processing on the content input by voice using a voice recognition device and narrows down the result of the voice recognition processing, the information depending on the notation is used as the additional information for narrowing down. , A program that causes a computer to execute a procedure for identifying a noun spoken by a user from a large number of search targets. 16. An individual requesting the user to voice-input an individual surname when the voice recognition processing is performed on the content input by the user using a voice recognition device and the result of the voice recognition processing is narrowed down. Last name input request procedure; Voice recognition procedure for recognizing voice input; Voice input request requesting the user to voice input at least one of the voice input personal last name notations A program for causing a computer to execute a procedure; a personal surname characteristic procedure for identifying the voice input personal surname based on the voice input personal surname and the voice input personal notation. 17. When voice recognition processing is performed on a content input by a user using a voice recognition device and the result of the voice recognition processing is narrowed down, information based on a difference in notation is used as additional information for narrowing down. , A computer-readable recording medium in which a program for causing a computer to execute a procedure for identifying a noun spoken by a user from a large number of retrieval targets is recorded. 18. An individual requesting the user to voice-input an individual surname when the voice-recognition device is used to perform voice-recognition processing on the content input by the user and to narrow down the result of the voice-recognition processing. First name input request procedure; Voice recognition procedure for recognizing voice input; Voice input request requesting the user to voice input at least one of the voice input personal last name notations A computer recording a program for causing a computer to execute a procedure; a personal surname characteristic procedure for specifying the voice-input personal surname based on the voice-input personal surname and the voice-input notation. A readable recording medium.