JP2004310692A - Failure resolution support device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To cope with colloqial expression of inputted voice to suppress generation of erroneous operation. <P>SOLUTION: A symptom sentence storage means 15 stores a symptom sentence described with failure information to a question assumed when cause of a failure is present, and a colloqial expression development means 16 converts the stored symptom sentence into a colloqial sentence. A voice recognition means 20 voice-recognizes the failure information by voice on the basis of a stochastic language model produced from the colloqial sentence converted by the colloquial expression development means 16. A cause candidate narrow-down means 21 extracts the corresponding symptom sentence obtained from a recognition result from the symptom sentence storage means 15, and narrows down a cause candidate on the basis of the extracted symptom sentence. A question selection means 22 selects a question for narrowing down the remaining cause candidate. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a failure resolution support device that inputs failure information and estimates the cause of a failure.
[0002]
[Prior art]
For example, as described in Patent Document 1, a conventional trouble solving support device includes an abnormal symptom input unit, an abnormality diagnosis information input unit, an inference unit, an output unit, and a function / component correspondence information storage unit. In the correspondence information storage means, each function of the abnormality diagnosis target device and the components involved in achieving the function are stored in association with each other, and the inference means stores the information based on the abnormality diagnosis information and the function / component correspondence information. Inferring abnormal diagnosis.
[0003]
Further, the trouble solving support device disclosed in Patent Literature 2 searches for necessary information from a predetermined database in response to a voice input from a user for the presented information. Means for classifying and storing into sets, narrowing means for narrowing information in the database according to voice input based on characteristics of each of the plurality of case sets, and a second method for the user based on the result of the narrowing. Means for prompting voice input, means for specifying a case matching the second voice input from a plurality of case sets, and presentation means for presenting predetermined information to the user based on the specified case. I have.
[0004]
[Patent Document 1]
JP-A-3-88036 (page 2, lower right column, means for solving the problem)
[Patent Document 2]
JP-A-11-306195 (paragraph 0022)
[0005]
Since the conventional trouble solving support device is configured as described above, in the case where a speech recognition device is combined with the trouble solving support device disclosed in Patent Document 1 or in the trouble solving support device described in Patent Document 2, There is a problem that it is not possible to cope with the spoken expression of the speech that has been generated, and a malfunction occurs.
[0006]
Further, when the quality of the input voice is poor or the performance of the voice recognition is insufficient, there is a problem that a malfunction due to a voice recognition error occurs.
[0007]
The present invention has been made in order to solve the above-described problems, and an object of the present invention is to provide a trouble solving support device that can cope with colloquial expression of input speech and can suppress occurrence of malfunction. I do.
[0008]
It is another object of the present invention to provide a failure resolution support device that can suppress the occurrence of a malfunction due to an error in speech recognition when the quality of input speech is poor or the performance of speech recognition is insufficient.
[0009]
[Means for Solving the Problems]
The problem solving support device according to the present invention includes a symptom sentence storage unit that stores a symptom sentence describing fault information for a question assumed when a cause of a fault exists, and converts the stored symptom sentence into a colloquial sentence. Spoken expression developing means, speech recognition means for recognizing speech based on a probabilistic language model created from colloquial sentence converted from input speech disability information by the spoken expression developing means, and corresponding symptoms from the recognition result A method comprising: extracting a sentence from a symptom sentence storage means; narrowing down a cause candidate based on the extracted symptom sentence; and a question selecting means for selecting and presenting a question for narrowing down the remaining cause candidates. It is.
[0010]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of the present invention will be described.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of a trouble solving support device according to Embodiment 1 of the present invention. This trouble solving support device inputs trouble information by a user's voice instead of inputting trouble information using a keyboard or the like.
[0011]
As shown in FIG. 1, this trouble solving support device includes a data input unit 11, a matrix creation unit 12, a matrix storage unit 13, a symptom sentence creation unit 14, a symptom sentence storage unit 15, a colloquial expression development unit 16, a corpus 17, The probabilistic language model creating unit 18, the probabilistic language model storage unit 19, the voice recognition unit 20, the cause candidate narrowing down unit 21, the question selecting unit 22, and the display unit 23.
[0012]
FIG. 2 is a block diagram showing an internal configuration of the colloquial expression expanding means 16 shown in FIG. As shown in FIG. 2, the colloquial expression expanding means 16 includes a morphological analysis means 31, a morphological information storage means 32, a colloquial substitution means 33 and a colloquial information storage means.
[0013]
Next, the operation will be described.
FIG. 3 is a flowchart showing a flow of a process of creating advance data in the failure solving support device according to the first embodiment of the present invention.
In step ST11, cause-question-answer data created by a data creator who is a specialist in failure diagnosis is input via the data input unit 11. The cause-question-answer data lists the cause of the failure and the question and answer expected when the cause exists.
[0014]
FIG. 4 is a diagram showing an example of cause-question-answer data. For example, as a question when there is a cause such as "blown fuse", a question "is the outlet normal?" And "is the power lamp lit?" If the answer to the question is "Normal" and the answer to the question "Is the power lamp on?" And the answer is "Off", it is possible that there is a cause of "Blown fuse". it can.
[0015]
These questions should be as common as possible for each cause. Further, it is not necessary to consider all questions and their answers for a certain cause, and there is no need to describe them.
[0016]
In step ST12 of FIG. 3, the matrix creating unit 12 creates a cause-question-answer matrix based on the input cause-question-answer data and stores the matrix in the matrix storage unit 13.
[0017]
FIG. 5 is a diagram showing an example of a cause-question-answer matrix created by the matrix creating means 12. In the cause-question-answer matrix, as shown in FIG. 5, causes and questions are arranged in a matrix format, and as the value of the matrix, each cause exists based on the input cause-question-answer data. The answer to the question assumed is entered.
[0018]
As shown in FIG. 5, the questions are numbered, for example, the first question is arranged so as to be in the first column of the matrix. Also, numbers are assigned to the causes, for example, the first cause is arranged so as to be on the first row. The answer which is the value of the matrix is digitized, and for example, “normal” is defined as “1”, and “abnormal” is defined as “2”. In FIG. 4, the answer to each question assumed when each cause exists is one, but when all the existing causes are collected, the answer to each assumed question is as shown in FIG. Be more than one.
[0019]
In FIG. 5, for example, the matrix creation unit 12 determines the value of the cause 1 (row) of “blown fuse” and the value of question 2 (column) of “whether the power lamp is turned on” (hatched portion in FIG. 5). The value "2" of "light off" is entered. Even if the answer number is the same, the meaning may differ if the question number is different. If the input of the cause-question-response data shown in FIG. 4 does not include a question and an answer for a certain cause, the matrix creating means 12 prepares the answer as a matrix value shown in FIG. A value quantifying that it is not performed, for example, “0” is entered.
[0020]
In step ST13 of FIG. 3, the symptom sentence creating means 14 creates a symptom sentence describing failure information from each question and the answer thereto, and creates the created symptom sentence and the question number and answer number when the symptom sentence was created. Is stored in the symptom sentence storage means 15. At this time, the symptom sentence creating means 14 creates symptom sentences for all possible answers to one question from the questions and answers described in the columns of the matrix in FIG. Remember.
[0021]
FIG. 6 is a diagram showing an example of the created symptom statement and original data. For example, the symptom sentence creating means 14 creates a symptom sentence “power lamp is lit” from question 2 “Is the power lamp on?” And answer 1 “light on”, and question 2 “the power lamp is on. "Is turned off" and the answer 2 "turn off" to create a symptom sentence "power lamp is off" and store it in symptom sentence storage means 15, respectively.
[0022]
In step ST14 of FIG. 3, the morphological analysis means 31 in the colloquial expression expanding means 16 shown in FIG. 2 reads out the symptom sentence stored in the symptom sentence storage means 15 and stores it in the morphological information storage means 32. The read symptom sentence is divided into morpheme units based on the morpheme information. For example, the morphological analysis unit 31 divides the symptom sentence “outlet is normal” into “outlet / wa / normal” in morpheme units. The specific method of morphological analysis is described in, for example, pages 91 to 98 of "Spoken Language Processing (Kenji Kita et al., Morikita Publishing Co., Ltd., 1996)".
[0023]
In step ST15, the colloquial expression replacing means 33 in the colloquial expression expanding means 16 shown in FIG. 2 refers to the colloquial expressions stored in the colloquial information storage means 34, and if there is a colloquial expression in each morpheme, the morpheme Is made into a colloquial sentence in which is replaced by the colloquial expression. For example, if a spoken word expression such as “OK” or “Is not strange” similar to “Normal” is stored in the spoken language information storage means 34 for the morpheme “Normal”, the spoken word replacing means 33 Create a colloquial sentence that is replaced with a spoken expression such as "outlet is OK" or "outlet is not strange".
[0024]
In step ST16, the colloquial substitution means 33 refers to the colloquial information storage means 34 and adds a morpheme at the end of the sentence such as "is" or "ne" or adds an interjection such as "erto" to a morpheme such as the beginning of a sentence. To create colloquial sentences in colloquial expressions. For example, the colloquial substitution means 33 creates a colloquial sentence in a colloquial expression such as "outlet is okay" or "um is not strange".
[0025]
In step ST17, the colloquial expression developing means 16 adds the colloquial sentence of the colloquial expression thus created to the corpus 17 as a sentence example. At this time, the colloquial expression developing means 16 stores information on which symptom sentence the colloquial sentence of the colloquial expression has been created.
[0026]
In step ST18, the probabilistic language model creating means 18 creates a probabilistic language model based on the colloquial sentences stored in the corpus 17, and stores it in the probabilistic language model storage means 19. Regarding the creation of a probabilistic language model based on a corpus and speech recognition using the same, for example, see “Probabilistic Language Model (Kenji Kita, University of Tokyo Press, 1999)” from page 23 to page 29, “Recognition error tendency”. Large Vocabulary Speech Recognition by Two-Stage Search Method Using Probability Model of Abe (Yoshiharu Abe et al., IEICE D-II Vol. J83-D-II No. 12, pp. 2545-pp. 2553) Has been described.
[0027]
FIG. 7 is a flowchart showing a flow of a fault solving process in the fault solving support device according to the first embodiment of the present invention. The user who uses this trouble solving support device may be a customer who has trouble information, an operator who has received a call from this customer, or the like.
[0028]
In step ST21, the voice recognition means 20 inputs the failure information by the voice of the user. In step ST22, the speech recognition unit 20 uses the prepared acoustic dictionary for recognition (not shown) and the probabilistic language model stored in the probabilistic language model storage unit 19 to input the input failure information. And outputs a speech recognition result.
[0029]
In step ST23, the cause candidate narrowing-down unit 21 compares the speech recognition result of the speech recognition unit 20 with the colloquial expression of the colloquial expression present in the corpus 17, and determines whether the colloquial expression of the colloquial expression in the corpus 17 matches or is similar. Detect sentences. In step ST <b> 24, the cause candidate narrowing down unit 21 detects a symptom sentence corresponding to the detected spoken language sentence from the symptom sentence storage unit 15.
[0030]
In step ST25, the cause candidate narrowing means 21 acquires the question number and the answer number as the original data of the detected symptom sentence from the symptom sentence storage means 15. For example, the speech recognition result "outlet is normal" is associated with a symptom sentence "outlet is normal", which is created from question number 1 and answer number 1 as shown in FIG. Similarly, the recognition result of “the power lamp is not turned on” is associated with a similar symptom sentence of “the power lamp is off”, which is created from question number 2 and answer number 2. The judgment that "the power lamp is not on" and "the power lamp is off" are similar by referring to the synonym data (not shown), and "not on" and "off" are synonyms. Is performed using the information.
[0031]
In step ST26, the cause candidate narrowing-down unit 21 uses the detected question number and answer number, refers to the cause-question-answer matrix stored in the matrix storage unit 13, and matches the input failure information. Eliminate candidate causes that do not. For example, when question number 1-answer number 1 is detected, causes other than those having values of 1 and 0 (0 is unanswered) in the first column (question number 1) are excluded from cause candidates. That is, in the example of FIG. 5, since the answer to the question 1 of the cause 2 is 2, the cause 2 is excluded from the cause candidates. When question number 2 to answer number 2 are detected, causes other than causes having values of 2 and 0 in the second column (question number 2) are excluded from cause candidates.
[0032]
In step ST27, the question selecting unit 22 searches the cause-question-answer matrix shown in FIG. 5 stored in the matrix storage unit 13 and selects the question that narrows the remaining cause candidates most efficiently. The question is selected such that there are many types of answers, and the remaining cause candidates for the type of answer are distributed equally, and the content of the question detected by the display means 23 is displayed to the user. I do.
[0033]
For example, in the example of FIG. 5, when the cause 2 is excluded from the cause candidates by the first failure information, the remaining cause candidates are the causes 1, 3, 4, and 5. Of the causes 1, 3, 4, and 5, the answer to Question 2 is 2 and 0, the answer to Question 3 is 0 and 1, and the answer to Question 4 is 0, 1 and 2. There are three types, the answer to Question 5 is 0, 2, and 3, the answer to Question 6 is 0, 2, and 1, and the answer to Question 7 is 0, 2, and 1. That is, questions 4, 5, 6, and 7 have many types of answers.
[0034]
Also, the number of cause candidates excluding cause 2 distributed to each answer is as follows: 2 answers 0 to question 4, 1 answer 1 and 1 answer, and 2 answers 0 to question 5 for question 4. 1 answer, 2 answers, 1 answer, 2 answers 0 for question 6, 1 answer, 1 answer, 2 answers 0 for question 7, There is one answer 2 and one answer 1. That is, the questions 4, 5, 6, and 7 have many types of answers, and the number of cause candidates distributed to each answer is a question that is narrowed down to the same degree of efficiency, and one of these questions is selected. .
[0035]
In step ST28, the voice recognition unit 20 performs voice recognition of the failure information by the user's voice, and in step ST29, the cause candidate narrowing down unit 21 selects a corresponding answer number in the same manner as in steps ST23 to ST25.
[0036]
Since the selected answer number is paired with the presented question number, in step ST30, the cause candidate narrowing means 21 matches the answer from the cause-question-answer matrix in the same manner as in step ST26. Eliminate causes that do not occur from the possible causes.
[0037]
In step ST31, the cause candidate narrowing-down means 21 determines whether or not the remaining cause candidates have been determined to be one. The coping method for the cause is displayed on the display means 23 and presented to the user. Here, the coping method for the cause is stored in a storage unit (not shown).
[0038]
In step ST31, when the remaining cause candidates are not determined to be one, in step ST33, the cause candidate narrowing means 21 determines whether or not each question is asked for the remaining cause candidates. By determining whether there are a plurality of types of answers, it is determined whether there is a question for narrowing down the cause candidates. If there is a question that narrows down the cause candidate, the process returns to step ST27 and repeats the above processing.
[0039]
In the above step ST33, if any question is asked for the remaining cause candidates and only a single answer is obtained, the cause candidate narrowing means 21 determines that there is no question for narrowing the cause candidates. In step ST34, the remaining cause candidates and the coping method for each cause candidate are displayed on the display means 23 and presented to the user.
[0040]
In the first embodiment, the cause candidate narrowing-down unit 21 detects a spoken language sentence in the corpus 17 similar to the speech recognition result, and detects a symptom sentence corresponding to the spoken language from the symptom sentence storage unit 15. However, the cause candidate narrowing-down unit 21 extracts a keyword related to the failure from the speech recognition result without detecting the spoken text in the corpus 17 and stores a symptom sentence including the keyword in the symptom sentence storage unit 15. May be detected.
[0041]
In the first embodiment, the cause candidate narrowing-down unit 21 removes typical spoken expressions such as “is” and “mas” from the speech recognition result without detecting the spoken language sentences in the corpus 17. After that, the corresponding symptom statement may be detected from the symptom statement storage unit 15.
[0042]
As described above, according to the first embodiment, the colloquial expression developing means 16 creates a colloquial sentence of a colloquial expression from a symptom sentence, and the speech recognition means 20 converts the input speech-based failure information from the colloquial sentence. Speech recognition is performed based on the created probabilistic language model, cause candidate narrowing means 21 extracts a corresponding symptom sentence from the recognition result, narrows the cause candidate based on the extracted symptom sentence, and causes question selecting means 22 to remain. By selecting and presenting the question for narrowing down the candidates, it is possible to cope with the spoken expression of the voice of the user, and it is possible to obtain an effect that occurrence of a malfunction can be suppressed.
[0043]
Embodiment 2 FIG.
FIG. 8 is a block diagram showing a configuration of a trouble solving support device according to a second embodiment of the present invention. This trouble-shooting support device performs a process of switching the target of voice recognition depending on the situation or switching from voice input to another input such as a keyboard when a customer and an operator exchange using a telephone or the like. is there.
[0044]
As shown in FIG. 8, the trouble solving support device includes a data input unit 11, a matrix creation unit 12, a matrix storage unit 13, a symptom sentence creation unit 14, a symptom sentence storage unit 15, a voice switching unit 24, and a speech recognition unit 20. , An alternative input unit 25, an input switching unit 26, a cause candidate narrowing down unit 21, a question selecting unit 22, and a display unit 23. Here, the voice switching unit 24 switches between the voice of the customer and the voice of the operator, and the input switching unit 26 switches between the input of the voice recognition result by the voice recognition unit 20 and the input of the alternative input unit 25 such as a keyboard.
[0045]
Next, the operation will be described.
FIG. 9 is a flowchart showing a flow of an input switching process in the failure solving support device according to the second embodiment of the present invention.
In step ST41, the control unit (not shown) sets the input switching unit 26 to the audio side, and in step ST42, the control unit sets the audio switching unit 24 to the customer side. In step ST43, the voice recognition means 20 recognizes the voice of the customer when the voice related to the fault information is input from the customer through a telephone or the like.
[0046]
In step ST44, the control unit calculates the reliability of the speech recognition result, for example, by comparing the likelihood of the speech recognition result with a predetermined threshold value. In step ST45, if the likelihood of the speech recognition result is larger than the predetermined threshold, in step ST46, it is determined that the recognition result of the customer's speech is sufficiently reliable, and the control unit keeps both the speech switching unit 24 and the input switching unit 26 as they are. Then, the recognition result of the customer's voice is used as an input and input to the cause candidate narrowing means 21.
[0047]
Subsequent processing shifts to step ST24 in FIG. 7 of the first embodiment, and the cause candidate narrowing-down unit 21 detects a symptom sentence corresponding to the speech recognition result from the symptom sentence storage unit 15, and proceeds to step ST25 and subsequent steps. Perform processing.
[0048]
If the likelihood of the voice recognition result is smaller than the predetermined threshold value in step ST45, the control unit sets the voice switching unit 24 on the operator side so that the voice of the operator is recognized instead of the voice of the customer.
[0049]
Generally, the voice of the customer comes through a telephone line or the like and is unclear and the sound quality is deteriorated. However, since the voice of the operator can be directly input through a microphone or the like, high quality can be expected. Further, by training the operator, it is possible to make the operator speak clearly so that the voice can be easily recognized. After switching to the operator side, the voice of the operator is recognized, so that the utterance content of the operator is to repeat the utterance of the customer as much as possible. For example, if a customer utters "printer characters are blurred," the operator repeats "the printer characters are blurred."
[0050]
In step ST48, the voice recognition means 20 recognizes the voice of the operator, and compares the likelihood of the voice recognition result with a predetermined threshold. In step ST49, if the likelihood of the voice recognition result is larger than the predetermined threshold, the process proceeds to step ST46, and the control unit determines that the recognition result of the voice of the operator is sufficiently reliable, and sets the voice switching unit 24 and the input switching unit 26. The recognition result of the voice of the operator who repeats the voice of the customer is used as an input and input to the cause candidate narrowing means 21. Subsequent processing moves to step ST24 in FIG. 7 of the first embodiment.
[0051]
In step ST49, if the likelihood of the voice recognition result is smaller than the predetermined threshold, the control means determines that the voice of the operator is not reliable, and in step ST50, switches the input switching means 26 to an alternative input such as a keyboard. Set on the means 25 side. In step ST51, the fault information from the alternative input means 25 is used as an input and is input to the cause candidate narrowing means 21. Subsequent processing moves to step ST24 in FIG. 7 of the first embodiment.
[0052]
In the second embodiment, a colloquial expression developing unit 16, a corpus 17, a stochastic language model creating unit 18, and a stochastic language model storage finger 19 shown in FIG. Alternatively, it may correspond to the spoken expression of the user's voice.
[0053]
That is, the speech recognition unit 20 recognizes the speech based on the probabilistic language model created from the colloquial sentence converted by the spoken expression development unit 16, and the cause candidate narrowing unit 21 A colloquial sentence in the corpus 17 similar to the speech recognition result may be detected, and a symptom sentence corresponding to the colloquial sentence may be detected from the symptom sentence storage means 15. Instead of detecting the spoken text in 17, a keyword related to the disorder may be extracted from the speech recognition result, and a symptom sentence including the keyword may be detected from the symptom sentence storage unit 15. After the speech means 21 removes typical spoken expressions such as “is” and “mas” from the speech recognition result without detecting the spoken language sentences in the corpus 17, the corresponding symptom sentences are displayed. May be detected from the sentence storage means 15.
[0054]
Further, in the second embodiment, the recognition result of the voice of the customer or the operator is used as an input in step ST46. However, the recognition result of the voice of the customer or the operator is used as the input, and both the recognition results match. Then, the recognition result may be input to the cause candidate narrowing-down unit 21.
[0055]
As described above, according to the second embodiment, if the voice of the customer is recognizable, the failure information is input by the voice of the customer, and if the voice of the customer is unclear and the voice of the operator is recognizable. If the voice of the operator who repeats the voice of the customer inputs the obstacle information, and if the voice of the customer and the operator is unclear, the obstacle information is input by an alternative input means such as a keyboard, so that the input voice is input. Even if the quality is poor, it is possible to suppress the occurrence of a malfunction due to a speech recognition error and to reduce the load of inputting fault information as a whole.
[0056]
Embodiment 3 FIG.
FIG. 10 is a block diagram showing a configuration of a trouble solving support device according to Embodiment 3 of the present invention. This trouble solving support device can suppress the occurrence of malfunction even if the recognition result includes a recognition error.
[0057]
As shown in FIG. 10, this trouble solving support apparatus includes a data input unit 11, a matrix creation unit 12, a matrix storage unit 13, a symptom sentence creation unit 14, a symptom sentence storage unit 15, a voice recognition unit 20, a narrowing down of a cause candidate. It comprises means 21, question selecting means 22, display means 23 and important keyword setting means 27. Here, the important keyword setting unit 27 sets an important keyword that characterizes failure information that is each answer to the question selected by the question selection unit 22.
[0058]
Next, the operation will be described.
FIG. 11 is a flowchart showing a flow of a fault solving process in the fault solving support device according to the third embodiment of the present invention. Steps ST61 and ST62 in FIG. 11 are the same as the processing in steps ST21 and ST22 in FIG. 7 of the first embodiment, and steps ST63 to ST66 are the same as the processing in steps ST24 to ST27.
[0059]
In step ST67, the important keyword setting means 27 characterizes the failure information that is each answer to the selected question and sets an important keyword that contributes to discrimination in each failure information. For example, in response to the question "What is the state of the LED?", The fault information has options of (1) "lit in red", (2) "flashing in red", and (3) "off". In this case, since the most important words for discriminating these are “lit”, “blinking”, and “extinguished”, “lit” corresponding to (1), “blinking” corresponding to (2), (3) ), The word “unlit” is set as an important keyword. Regarding the selection of an important keyword, for example, a word that is not common to each piece of failure information may be set as an important keyword. Alternatively, a skilled person may create the data in advance. The number of important keywords is not limited to one for each piece of failure information, and a plurality of important keywords may be set.
[0060]
In step ST68, the voice recognition means 20 recognizes the failure information by the voice of the user and outputs a voice recognition result. In step ST69, the cause candidate narrowing means 21 performs a morphological analysis on the speech recognition result, and searches for an important keyword set by the important keyword setting means 27.
[0061]
In step ST70, the cause candidate narrowing means 21 checks whether or not the set important keyword is included in the speech recognition result, and if there is an important keyword, selects failure information corresponding to the important keyword. For example, when the user utters “lit in red”, but the recognition result is “arc lighting” due to poor performance of the voice recognition means 20, the important keyword “lighting” is present. The candidate narrowing means 21 selects the failure information (1) “lights up in red”. When there are a plurality of important keywords, a fault is assigned to each important keyword so as to give the highest total score. Further, the cause candidate narrowing-down unit 21 extracts a symptom sentence corresponding to the selected failure information from the symptom sentence storage unit 15, and selects an answer number from the extracted symptom sentence.
[0062]
The processing of steps ST71 to ST75 is the same as the processing of steps ST30 to ST34 in FIG. 7 of the first embodiment.
[0063]
In the third embodiment, a colloquial expression developing unit 16, a corpus 17, a stochastic language model creating unit 18, and a stochastic language model storage unit 19 shown in FIG. 1 of the first embodiment are provided. Alternatively, it may correspond to the spoken expression of the user's voice.
[0064]
That is, the speech recognition unit 20 recognizes the speech based on the probabilistic language model created from the colloquial sentence converted by the spoken expression development unit 16, and the cause candidate narrowing unit 21 A colloquial sentence in the corpus 17 similar to the selected failure information may be detected, and a symptom sentence corresponding to the colloquial sentence may be detected from the symptom sentence storage means 15. Instead of detecting the spoken text in the corpus 17, a keyword related to the failure may be extracted from the selected failure information, and a symptom sentence including the keyword may be detected from the symptom sentence storage unit 15. The candidate narrowing-down unit 21 removes standard spoken expressions such as “is” and “mas” from the selected failure information without detecting spoken sentences in the corpus 17, and then responds. May be detected symptoms statement from symptoms sentence storage means 15.
[0065]
Further, in the third embodiment, the cause is narrowed down using the cause-question-answer matrix. However, even when the cause is narrowed down and specified by the tree-structured question-answer data, the reliability is improved by the same processing. It is possible to improve.
[0066]
As described above, according to the third embodiment, even when the performance of speech recognition is insufficient, it is possible to correctly process an important keyword that can discriminate each piece of failure information, and improve reliability. The effect that it can be obtained is obtained.
[0067]
In the first to third embodiments described above, it is needless to say that each means can be configured by any of hardware and software. When configured by software, the program may be a program recorded on a recording medium or a program downloaded via a communication medium.
[0068]
【The invention's effect】
As described above, according to the present invention, a symptom sentence storage unit that stores a symptom sentence that describes failure information for a question assumed when a cause of a failure exists, and the stored symptom sentence is converted into a spoken sentence. A speech recognition unit that converts the spoken language expression to be converted, a speech recognition unit that performs speech recognition on the basis of a probabilistic language model created from the spoken language sentence converted by the spoken expression development unit and converts the obstacle information by the input speech, and The system comprises: a cause candidate extracting unit that extracts a symptom sentence from the symptom sentence storage unit and narrows down the cause candidate based on the extracted symptom sentence; and a question selecting unit that selects and presents a question for narrowing down the remaining cause candidates. Thereby, it is possible to cope with the spoken expression of the user's voice, and it is possible to suppress the occurrence of a malfunction.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a failure solution support device according to a first embodiment of the present invention.
FIG. 2 is a block diagram showing an internal configuration of a colloquial expression developing unit of the trouble solving support device according to the first embodiment of the present invention.
FIG. 3 is a flowchart showing a flow of a process of creating advance data in the failure solving support device according to the first embodiment of the present invention.
FIG. 4 is a diagram showing an example of cause-question-answer data input to the failure resolution support device according to the first embodiment of the present invention.
FIG. 5 is a diagram showing an example of a cause-question-answer matrix created by the fault solution support device according to the first embodiment of the present invention.
FIG. 6 is a diagram showing an example of a symptom sentence and original data created by the trouble solving support device according to the first embodiment of the present invention.
FIG. 7 is a flowchart showing a flow of a fault solving process in the fault solving support device according to the first embodiment of the present invention.
FIG. 8 is a block diagram showing a configuration of a failure solution support device according to a second embodiment of the present invention.
FIG. 9 is a flowchart showing a flow of an input switching process in the failure solving support device according to the second embodiment of the present invention.
FIG. 10 is a block diagram showing a configuration of a trouble solving support device according to a third embodiment of the present invention.
FIG. 11 is a flowchart showing a flow of a fault solving process in the fault solving support device according to the third embodiment of the present invention.
[Explanation of symbols]
11 data input means, 12 matrix creation means, 13 matrix storage means, 14 symptom sentence creation means, 15 symptom sentence storage means, 16 colloquial expression development means, 17 corpus, 18 stochastic language model creation means, 19 stochastic language model storage Means, 20 voice recognition means, 21 cause candidate narrowing means, 22 question selection means, 23 display means, 24 voice switching means, 25 alternative input means, 26 input switching means, 27 important keyword setting means, 31 morphological analysis means, 32 Morphological information storage means, 33 colloquial substitution means, 34 colloquial information storage means.

Claims (12)

In a fault resolution support device that inputs fault information and estimates the cause of the fault,
Symptom statement storage means for storing a symptom statement describing failure information for a question assumed when a cause of the failure exists,
Colloquial expression expansion means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
Speech recognition means for recognizing speech based on a probabilistic language model created from the spoken language sentence converted from the obstacle information by the input speech by the spoken expression development means,
Cause candidate narrowing means for extracting a corresponding symptom sentence from the symptom sentence storage means from the recognition result by the voice recognition means and narrowing down a cause candidate based on the extracted symptom sentence,
A problem solving support device comprising: a question selecting unit that selects and presents a question for narrowing down the remaining cause candidates as a result of the narrowing down by the cause candidate narrowing down unit. 2. The cause candidate narrowing means detects a spoken language sentence similar to the recognition result by the voice recognition means, and extracts a symptom sentence corresponding to the detected spoken language sentence from the symptom sentence storage means. Trouble solving support device. 2. The trouble solving support device according to claim 1, wherein the cause candidate narrowing means extracts a symptom sentence including a keyword relating to the obstacle from the result of recognition by the voice recognition means from the symptom sentence storage means. 2. The trouble solving support apparatus according to claim 1, wherein the cause candidate narrowing-down unit removes the spoken expression from the recognition result by the voice recognition unit and extracts a corresponding symptom sentence from the symptom sentence storage unit. In a fault resolution support device that inputs fault information and estimates the cause of the fault,
Voice switching means for switching between a voice from a customer holding the failure information and a voice from an operator who repeats the failure information by contact from the customer;
Voice recognition means for voice recognition of the fault information by voice switched by the voice switching means,
An alternative input means for inputting fault information by means other than voice;
A recognition result of the voice recognition means, and input switching means for switching failure information from the alternative input means,
Symptom statement storage means for storing a symptom statement describing failure information for a question assumed when a cause of the failure exists,
Extracting the corresponding symptom sentence from the symptom sentence storage means from the recognition result or the failure information switched by the input switching means, and narrowing down the cause candidate based on the extracted symptom sentence,
A problem solving support device comprising: a question selecting unit that selects and presents a question for narrowing down the remaining cause candidates as a result of the narrowing down by the cause candidate narrowing down unit. A colloquial expression developing means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
The speech recognition means performs speech recognition based on the probabilistic language model created from the spoken language sentence converted by the spoken expression development means, and outputs the obstacle information by the input speech,
6. The method according to claim 5, wherein the cause candidate narrowing-down unit detects a colloquial sentence similar to the recognition result by the voice recognition unit, and extracts a symptom sentence corresponding to the detected colloquial sentence from the symptom sentence storage unit. The trouble solving support device according to the above. A colloquial expression developing means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
The speech recognition means performs speech recognition based on the probabilistic language model created from the spoken language sentence converted by the spoken expression development means, and outputs the obstacle information by the input speech,
6. The trouble solving support apparatus according to claim 5, wherein the cause candidate narrowing means extracts a symptom sentence including a keyword relating to the obstacle from the result of recognition by the voice recognition means from the symptom sentence storage means. A colloquial expression developing means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
The speech recognition means performs speech recognition based on the probabilistic language model created from the spoken language sentence converted by the spoken expression development means, and outputs the obstacle information by the input speech,
6. The trouble solving support apparatus according to claim 5, wherein the cause candidate narrowing-down unit removes the spoken expression from the recognition result by the voice recognition unit and extracts a corresponding symptom sentence from the symptom sentence storage unit. In a fault resolution support device that inputs fault information and estimates the cause of the fault,
Symptom statement storage means for storing a symptom statement describing failure information for a question assumed when a cause of the failure exists,
Voice recognition means for voice-recognizing the failure information by the input voice;
Cause candidate narrowing means for extracting a corresponding symptom sentence from the symptom sentence storage means from the recognition result by the voice recognition means and narrowing down a cause candidate based on the extracted symptom sentence,
Question selection means for selecting and presenting a question for narrowing down the remaining cause candidates as a result of being narrowed down by the cause candidate narrowing down means,
Important keyword setting means for setting an important keyword that characterizes each failure information assumed for the question selected by the question selecting means,
The cause candidate narrowing means searches the important keyword set by the important keyword setting means from the recognition result by the voice recognition means, selects fault information, and extracts a corresponding symptom sentence. Failure resolution support device. A colloquial expression developing means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
The speech recognition means performs speech recognition based on the probabilistic language model created from the spoken language sentence converted by the spoken expression development means, and outputs the obstacle information by the input speech,
10. The failure candidate according to claim 9, wherein the cause candidate narrowing-down unit detects a spoken language sentence similar to the selected failure information, and extracts a symptom sentence corresponding to the detected spoken language sentence from the symptom sentence storage unit. Solution support device. A colloquial expression developing means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
The speech recognition means performs speech recognition based on the probabilistic language model created from the spoken language sentence converted by the spoken expression development means, and outputs the obstacle information by the input speech,
10. The trouble solving support device according to claim 9, wherein the cause candidate narrowing means extracts a symptom sentence including a keyword relating to the fault from the selected fault information from the symptom sentence storage means. A colloquial expression developing means for converting the symptom sentence stored in the symptom sentence storage means into a colloquial sentence,
The speech recognition means performs speech recognition based on the probabilistic language model created from the spoken language sentence converted by the spoken expression development means, and outputs the obstacle information by the input speech,
10. The trouble solving support device according to claim 9, wherein the cause candidate narrowing down means removes the spoken expression from the selected trouble information and extracts a corresponding symptom sentence from the symptom sentence storage means.

Images