JP3557467B2 - Method and apparatus for recognizing voice data - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a system for performing operations such as operation, maintenance, and inspection at a plant or a factory using a computer, and in particular, to voice data suitable for improving the recognition rate of voice data when voice data is input. The present invention relates to a recognition method and an apparatus thereof.
[0002]
[Prior art]
2. Description of the Related Art In recent years, use of a voice recognition device for operation, maintenance, and inspection of a power plant has been promoted. This aims to simplify the inspection work and the like by inputting and recording, for example, the inspection result by voice. Although practical use of speech recognition has been promoted in this way, the speech recognition rate does not always have sufficient performance. To make the system easy to use with good man-machine properties, it is essential to improve the recognition rate of speech recognition.
[0003]
In conventional speech recognition, an input speech pattern created by extracting a feature parameter from an input speech signal is compared with a standard pattern stored in a dictionary in advance to obtain a similarity. The highest word is the recognition result. However, in this method, words having similar pronunciations are likely to be erroneously recognized. In addition, particularly for short words such as numbers, there is a small amount of information to be compared with the standard pattern, so that there is a drawback that erroneous recognition is likely. In operation, maintenance, and inspection of a plant or the like, there are many similar system / device names. In many cases, numerical values such as temperature and flow rate are input. For this reason, in order to apply the speech recognition technology to plant operation or the like, it is necessary to enhance the function of preventing erroneous recognition of similar words and numerical values.
[0004]
As a conventional method for reducing erroneous recognition, for example, a method described in JP-A-61-122781 is known. In this conventional speech recognition method, a candidate word is selected by comparing an input signal with a standard pattern, and a recognition word is determined by taking into account the degree of language suitability obtained based on the frequency of use.
[0005]
[Problems to be solved by the invention]
However, this conventional speech recognition method cannot be applied to plant inspection and the like. This is because, depending on the inspector and the inspection date, the inspection target and the situation change dynamically, so the frequency of use of the word also changes dynamically. This is because the type and range are changed. That is, in the above-described conventional technology, the recognition rate of voice recognition cannot be improved when the target or the situation changes dynamically, such as in a plant inspection.
[0006]
SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech data recognition method and apparatus capable of improving the recognition rate of speech recognition irrespective of a target to which speech recognition is applied and a change in situation.
[0007]
[Means for Solving the Problems]
The object is to provide a speech data recognition method for comparing an input speech pattern and a standard speech pattern with respect to a word input by speech and determining a candidate word based on a similarity between the two. When the candidate word and the candidate rank of the candidate word are obtained by comparing the input voice pattern and the standard voice pattern, and the candidate word is numerical data, the characteristics and the characteristics of the object related to the numerical data are obtained. A process of changing the candidate order of the candidate word is performed in accordance with the relationship with the candidate word. This is achieved by determining the candidate order of the candidate words.
[0008]
The above purpose is also A relational expression relating to a numerical range that can be taken by the input numerical data is predetermined, When numerical data is input by voice and a plurality of candidate words for the input voice are determined by comparison with a standard voice pattern, the candidate word is selected from among the plurality of candidate words. Relational expression Is determined by determining a candidate word that satisfies the above as a high-order candidate rank.
[0009]
The above object is also provided that, when numerical data for the same thing that has been input and speech recognized in the past is input by speech and a plurality of candidate words are determined by comparison with a standard voice pattern, the candidate word is selected from among the plurality of candidate words in the past. This is achieved by determining numerical data that is close to the numerical data for which speech recognition has been performed, as a high-order candidate ranking.
[0010]
The above object is also achieved in that, when measurement data for a target whose standard value is known in advance is input as speech and a plurality of candidate words are determined by comparison with a standard voice pattern, the standard value is selected from the plurality of candidate words. This is achieved by determining a candidate word representing close measurement data as a higher candidate rank.
[0011]
The above purpose is also Objects related to the work performed by the voice entrant Terms that are likely to be spoken based on As preferred words If a plurality of candidate words are determined in advance and are determined based on a speech recognition result using a standard speech pattern for the input speech, Based on the object, the priority word This is achieved by determining a candidate word that matches with as the top candidate rank.
[0012]
The above object is also achieved by, when a plurality of candidate words are determined based on a speech recognition result, determining a candidate order of the plurality of candidate words based on characteristics of a speaker.
[0013]
According to the present invention, for example, when the temperature data is inputted as a voice as a numerical value and “215” and “115” are determined as the recognition result, the conditional expression to be satisfied by the temperature data is 100 degrees or more and less than 120 degrees. When it is determined separately, the recognition rate is improved by determining “115” as the candidate higher rank.
[0014]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a configuration diagram of a speech data recognition device according to a first embodiment of the present invention. The voice data recognition device according to the present embodiment includes a recognition device main body 1, a portable input / display device 2 wirelessly connected to the main body 1, and an external input device 3 attached to the main body 1. The plant database 19 of the main unit 1 of the speech recognition device is loaded with various process data from the plant 7 via the control computer 6.
[0015]
The input / display device 2 is carried by an inspector, inspects various parts of the plant, inputs the inspection results by voice, and displays the inspection results (speech recognition results performed by the main unit 1 on the input voice). It comprises a microphone 27, a portable display 29, and a transmitting / receiving unit (mobile station) 28 communicating with the recognition device main body 1.
[0016]
The recognition device body 1 is recognized by a transmission / reception unit (fixed station) 11, a voice recognition unit 12 for converting a voice signal input from the microphone 27 and captured through the transmission / reception unit 28/11 into a character signal, and a voice recognition unit 12. A speech recognition dictionary 13 storing a standard pattern used when the input / output is performed, an input / output signal of a transmission / reception unit (fixed station) 11, an input signal from an external input device 3, an output signal to an external display device 4, and a About an input / output control unit 14 for controlling an output signal, an inspection record management unit 15 for outputting an inspection record table and recording an inspection result, an inspection record storage unit 16 for storing an inspection record, and an audio signal of numerical data. Numerical data determination unit 17 that changes the candidate ranking of candidate words as described in detail below, and a sensor installation information storage unit that stores information on sensors to be checked and process amounts 8, consisting of the plant database 19 for storing the process variable taken in through the control computer 6.
[0017]
FIG. 2 is a flowchart illustrating a processing procedure when the inspection result is stored in the inspection record storage unit 16. When the inspector inputs the inspection result by voice from the microphone 27 of the input / display device 2, the voice signal is converted into an electric signal and then transferred to the transmitting / receiving unit 11 of the recognition device main body 1 via the transmitting / receiving unit 28. (Step 1). The transmission / reception section (fixed station) 11 sends the transferred signal to the input / output control section 14. If the signal from the transmission / reception section 11 is a signal based on a voice input, the input / output control section 14 transmits the signal to the voice recognition section 12. Send to
[0018]
The voice recognition unit 12 performs a recognition process using the voice recognition dictionary 13. The candidate words and the candidate ranks as a result of the recognition are output to the inspection record management unit 15 (step 2). The candidate word refers to a word (usually, a plurality of words) in which the similarity between a speech input signal and a standard pattern registered in a speech dictionary is equal to or greater than a preset value in the recognition processing. The candidate ranking is a number that is assigned an order from a candidate word in a descending order of similarity. The one with the highest candidate rank is set as the first candidate, and is processed as the second candidate, the third candidate, and so on.
[0019]
The inspection record management unit 15 records the inspection result in the inspection record storage unit 16 according to the signal of the recognition result of the voice recognition unit 12. However, when the candidate word sent from the voice recognition unit 12 is numerical data, the inspection record storage unit 16 outputs the candidate word and the candidate rank to the numerical data determination unit 17. At this time, the numerical data determination unit 17 refers to the sensor installation information storage unit 18 and the plant database 19, and changes the candidate order of the candidate words as described in detail later. The sensor installation information storage unit 18 stores a relational expression between an inspection result and a process amount and an upper limit value and a lower limit value of the inspection result for each inspection target and inspection item. Based on this information, the candidate order is changed by comparing the value of the process amount acquired from the plant database 19 with the candidate word. After changing the candidate order of the candidate words, the numerical data determination unit 17 outputs the candidate words to the inspection record management unit 15 (Step 4). The inspection record management unit 15 stores the inspection record in the inspection record storage unit 15 using the word having the highest candidate rank among the candidate words sent from the numerical data determination unit 17 (Step 5).
[0020]
FIG. 3 is a diagram illustrating a configuration example of the inspection data stored in the inspection record storage unit 16. The inspection record storage unit 16 stores data on inspection targets, inspection items, and inspection results. Of these, the inspection target and the inspection items are determined in advance, and the item data and the like are already stored. The inspection record management unit 15 stores the data of the inspection result in each column of the corresponding inspection target and inspection item.
[0021]
The recognition device body 1 according to the present embodiment, when numerical data is input by voice from the portable input / display device 2, applies a process amount and an inspection item to a plurality of candidate words recognized by the voice recognition unit 12. The candidate ranking is changed using such information. The principle of changing the candidate order will be described with reference to FIG.
[0022]
In FIG. 4, 31 is a water supply pipe. 32, 33, and 34 are detectors for measuring the feedwater temperature. The latest value of the feedwater temperature automatically measured by each of the detectors 32 and 34 is stored in the plant database as a process amount. These feedwater temperatures are T1 and T2, respectively. Here, the inspector reads the indicated value of the feedwater temperature T measured by the detector 33 provided between the upstream detector 34 and the downstream detector 32, and reads the numerical value using the microphone 27. Enter
[0023]
T1 <T <T2 is established by the relationship between the installation positions of the detectors shown in FIG. 4 (assuming that heat is not input to the water supply pipe 31). Therefore, in the present embodiment, when there is a plurality of candidate words detected by the voice recognition unit 12 by comparing the input voice pattern with the standard pattern, the candidate order is changed based on this conditional expression. For example, when T1 is “100”, T2 is “120”, and the candidate word, that is, the value corresponding to T is “215” (first candidate) and “115” (second candidate), it is the first candidate. “215” does not satisfy 100 <T <120. Therefore, the first candidate is changed to “115” that satisfies the above conditional expression, and the second candidate is changed to “215”.
[0024]
FIG. 5 is a flowchart illustrating a processing procedure in the numerical data determination unit 17. The numerical data determination unit 17 operates when a numerical value is input from the inspection record management unit 15 (Step 11). The input data is information on an inspection target, an inspection item, a candidate word, and a candidate rank. The numerical data determination unit 17 acquires information on the input inspection target and inspection items from the sensor installation information storage unit 18 (step 12).
[0025]
FIG. 6 is a configuration diagram of the sensor installation information storage unit 18. The sensor installation information storage unit 18 is composed of data of “inspection target”, “inspection item”, “reference process amount”, and “condition”. The “reference process amount” is a process amount to be compared with the inspection result. The "condition" stores an upper limit value and a lower limit value of the inspection result, a relational expression established between the inspection result and the reference process amount, and the like.
[0026]
The numerical data determination unit 17 acquires “reference process amount” and “condition” data based on the inspection target and the inspection item input from the sensor installation information storage unit 18, and then obtains a plant database 19 (FIG. 1). The acquisition value corresponding to the above-mentioned reference process amount is acquired from (step 13 in FIG. 5). FIG. 7 is a configuration diagram of the plant database. The plant database 19 includes data of “system name”, “process amount”, and “captured value”.
[0027]
Next, the numerical data determination unit 17 collates the acquired value of the process amount with the condition acquired from the sensor installation information storage unit 18 in FIG. 6 to determine whether each candidate word satisfies the conditional expression. I do. Then, among the candidate words, the candidate rank of the candidate word that does not satisfy the conditional expression is set to the lowest (step 14). After changing the candidate order of the candidate words in this way, the numerical data determination unit 17 outputs the candidate words and the candidate orders to the inspection record management unit 15 (step 15).
[0028]
For example, “Ichi Ichigo” is input by voice for the inspection target “water supply pipe” and the inspection item “temperature”, and the candidate word “215” (first candidate) is obtained as a voice recognition result for the input. And “115” (second candidate) are input to the numerical data determination unit 17. In this case, the numerical data determination unit 17 acquires the reference process amount and the condition corresponding to the inspection target “water supply pipe” and the inspection item “temperature” from the sensor installation information unit. In the sensor installation information storage unit shown in FIG. 6, the reference process amounts are “A pipe temperature” and “B pipe temperature”. Therefore, the numerical data determination unit 17 acquires “A pipe temperature” and “B pipe temperature” from the plant database 19. In the information of the plant database in FIG. 7, “A pipe temperature” is 100 ° C., and “B pipe temperature” is 120 ° C.
[0029]
Next, the numerical data determination unit 17 collates the process amount with the condition acquired from the sensor installation information storage unit 18. This condition is T1 (A pipe temperature) <T <T2 (B pipe temperature). Therefore, a conditional expression of 100 <T <120 is obtained. In this case, since the candidate words are “215” (first candidate) and “115” (second candidate), the first candidate “215” does not satisfy this conditional expression. Therefore, the numerical data determination unit 17 changes “115” satisfying the conditional expression to the first candidate, and changes “215” to the second candidate.
[0030]
FIG. 8 is a flowchart illustrating a processing procedure of the inspection record management unit 15. The inspection record management unit 15 operates when there is input data from the voice recognition unit 12 (Step 21). The input data is a candidate word corresponding to a check target, a check item, a check result, and a candidate rank thereof. When an inspection object is input (step 22), an inspection record table corresponding to the inspection object is output to the input / output control unit 14 (step 30), and input of an inspection item is waited for (step 31).
[0031]
When the check item is input (step 23), the check item is output to the input / output control unit 14 (step 32), and the input of the check result is awaited (step 33). By outputting the inspection record table and the inspection items to the input / output control unit 14, these are displayed on the portable display 29 (FIG. 1). At the time of input of the inspection target and the inspection item, no numerical data is input. Therefore, the word having the highest candidate rank among the candidate words is used for these processes.
[0032]
Next, when the inspection result is input, the processing of candidate words for the inspection result is different between numeric data and character data. If the candidate word is character data, the character having the highest candidate rank among the candidate words is output to the input / output control unit 14 (step 28) and stored in the inspection record storage unit 16 (step 29). If the candidate word is numerical data (step 25), the “inspection target”, “inspection item”, “candidate word”, and “candidate ranking” are output to the numerical data determination unit 17 (step 26).
[0033]
The numerical data determination unit 17 compares each candidate word of the numerical data with the conditional expression, changes the candidate word candidate order by the above-described processing as necessary, and sends the candidate word and its The candidate ranking is output (step 27).
[0034]
The inspection record management unit 15 stores the word with the highest candidate rank among the candidate words input from the numerical data determination unit 17 in the inspection record storage unit 16 (step 28), and outputs this to the input / output control unit 14. (Step 29). When the inspector inputs a word other than the inspection result by voice while the system requests the input of the inspection result (step 24), the inspection record management unit 15 enters an error message displayed on the portable display 29. Output to the output control unit 14 (step 34).
[0035]
FIG. 9 is a flowchart illustrating a processing procedure of the voice recognition unit 12. The voice recognition unit 12 operates when a voice signal is input from the input / output control unit 14. When a voice signal is input (step 36), the similarity is determined by comparing with a standard pattern registered in the voice recognition dictionary 13 (step 37). All the words corresponding to the standard pattern whose similarity is higher than a preset value are set as candidate words, and the candidate words are ranked in the order of higher similarity (step 38). Then, these candidate words and candidate ranks are output to the inspection record management unit 15 (step 39).
[0036]
FIG. 10 is a flowchart illustrating a processing procedure of the input / output control unit 14. When there is an input from the inspection record management unit 15 (step 41), the input / output control unit 14 outputs this signal to the transmission / reception unit (fixed station) 11 (step 44). This is for displaying the inspection record table or the words input by voice on the portable display 29. When there is an input from the transmission / reception unit (fixed station) 11 (step 42), this signal is output to the voice recognition unit 12 to convert the voice signal into a character signal (step 45). When an inspection target is input using the external input device 3 (step 43), this is output to the inspection record management unit 15. At this time, the inspection record management unit 15 outputs the input inspection record table of the inspection target to the input / output control unit 14. The input / output control unit 14 outputs an inspection record table to the printer 5 or the external display device 4 according to the input request (step 46).
[0037]
FIG. 11 is a diagram showing a display screen example of a portable display showing an operation example of the present invention. The portable display 29 is used to check the inspection record input by voice. The initial screen (FIG. 7A) is in a state of waiting for an input of an inspection target. The highlight indicating the input waiting state is lit at the input location to be inspected. In the state of the initial screen, the inspection target is input as an audio signal from the microphone 27. When the inspector speaks "water supply piping", an inspection record table of the water supply piping is displayed on the portable display 29 on the basis of the result of the above-mentioned processing as shown in FIG. Similarly, when the inspector speaks “temperature”, the state is awaiting the input of the inspection result for the temperature. At the temperature input location, as shown in FIG. 3C, a highlight indicating the input waiting state is turned on. When the inspector utters “115”, the first candidate “115” by the above-described processing is displayed in the item of the temperature on the portable display 29 as shown in FIG. If the input value or the like is erroneously recognized in the course of these processes, the next candidate is displayed by speaking “next”. If the user wants to correct the input value, the user can re-enter the value after saying "correction". The inspection result input by the above processing is displayed on the portable display 29 and also recorded in the inspection record storage unit 16.
[0038]
According to the present embodiment, when the inspection result of the numerical data is input by voice, the candidate ranking of the candidate word based on the voice recognition result is related to the configuration and function of the facility such as the current value of the process amount and the installation position of the sensor. Since the determination is made based on the data, the recognition rate when numerical data is input by voice is improved. Incidentally, even if a trend value of the process amount is used instead of the current value of the process amount, the recognition efficiency of speech recognition can be improved by the same processing. In addition, the present invention improves the recognition rate of voice recognition of numerical data by simply adding a device for evaluating the validity of numerical data to a conventional voice recognition device, so that the system of the present invention can be easily configured. Can be. Further, the present invention refers to a plant database that stores the latest values related to the plant state, even when the inspection result that changes sequentially according to the state of the plant is input by voice, so that the input data is always stored regardless of the state change of the plant. Can evaluate validity.
[0039]
FIG. 12 is a configuration diagram of a speech data recognition device according to the second embodiment of the present invention. The recognition apparatus of the present embodiment is different from the recognition apparatus shown in FIG. 1 in that the plant database 19 and the sensor installation information storage unit 18 are deleted, and the inspection record storage unit 16 is also connected to the numerical data determination unit 17. different. That is, the configuration of the inspection record storage unit 16 and the processing in the numerical data determination unit 17 are different.
[0040]
FIG. 13 is a configuration diagram of inspection data stored in the inspection record storage unit 16 according to the second embodiment. The inspection record storage unit 16 includes “inspection target”, “inspection item”, “inspection result”, and “(previous value) inspection result”. Each data of the inspection object, the inspection item, and the (previous value) inspection result is stored in advance. The inspection record management unit 16 stores the current inspection result in the corresponding inspection target and inspection item.
[0041]
FIG. 14 is a flowchart illustrating a processing procedure in the numerical data determination unit 17 according to the second embodiment. The numerical data determination unit 17 operates when there is an input from the inspection record management unit 15 (step 51). The input data is a check target, a check item, a candidate word, and a candidate rank. First, the numerical data determination unit 17 acquires the inspection result of the previous value for the input inspection target and inspection item from the inspection record storage unit 16 (step 52).
[0042]
Next, the difference between the numerical value of the candidate word, which is the recognition result of the voice data with respect to the current inspection result, and the previous value is calculated. If the difference is larger than the preset value, the candidate rank of this candidate word is reset to the lowest value (step 53). After changing the candidate order of each candidate word, the numerical data determination unit 17 outputs the candidate word and its candidate order to the inspection record management unit 15 (step 54).
[0043]
For example, a process when the inspection target “water supply pipe”, the inspection item “temperature”, and the candidate words “215” (first candidate) and “115” (second candidate) are input to the numerical data determination unit 17 will be described. . First, the numerical data determination unit 17 acquires the previous value from the inspection record storage unit 16. In the inspection record storage unit 16 illustrated in FIG. 13, the previous value corresponding to the inspection target “water supply pipe” and the inspection item “temperature” is “113”. Next, a difference between each of the candidate words “215” and “115” and the previous value “113” is calculated. At this time, if a preset value that the difference from the previous value is within 10% is provided in advance, the first candidate “215” does not satisfy this condition. Therefore, the numerical data determination unit 17 changes “115” to the first candidate and changes “215” to the second candidate.
[0044]
According to this embodiment, when the check result of the numerical data is input by voice, the candidate ranking of the candidate word is determined with reference to the previous value, so that the recognition rate can be improved. Thus, in the case where the same part is regularly inspected, the possibility of erroneous recognition is reduced by using the previous value, the trouble such as re-input is reduced, and the burden on the input person can be reduced. In the present embodiment, the comparison target of the candidate word is only the previous value. However, even if the average value of the past inspection results is used, the similar processing can improve the recognition rate. Further, in the present embodiment, the device can be easily configured simply by adding a device for storing the previous value of the inspection result and a device for evaluating the validity of the numerical data to the conventional voice input system. Development can be performed easily.
[0045]
Further, in the present embodiment, since a function of obtaining a difference between the input inspection result and the previous value corresponding thereto is provided, by adding a function of displaying this value to the user, the user can reduce the deviation from the previous value. It is possible to confirm, and it is possible to early detect a plant abnormality.
[0046]
The recognition device according to the present embodiment can be applied not only to plant inspection but also to various fields. For example, consider a situation in which numerical data arranged in a monotonically decreasing order from a large value to a small value is input by voice. In this case, since the numerical data input by voice is always smaller than the previous value, the candidate word is compared with the previous value, and if the candidate word is larger than the previous value, the candidate rank is lowered or deleted from the candidate word. Good. When numerical data to be input by voice has some regularity as in this example, the recognition rate of voice recognition can be improved by using the same technology as the apparatus of the present embodiment.
[0047]
FIG. 15 is a configuration diagram of a speech data recognition device according to the third embodiment of the present invention. In the second embodiment of FIG. 12, the inspection result of the previous value is input to the numerical data determination unit 17 and compared with the comparison result. However, in this embodiment, the standard value corresponding to the inspection item is smaller than that of the second embodiment. The only difference is that a stored data standard value storage unit 20 is newly provided and the stored data is input to the numerical data determination unit 17.
[0048]
FIG. 16 is a flowchart illustrating a processing procedure in the numerical data determination unit 17 according to the third embodiment. The numerical data determination unit 17 operates when there is an input from the inspection record management unit 15 (step 56). The input data is a check target, a check item, a candidate word, and a candidate rank. First, the numerical data determination unit 17 acquires standard values for the input inspection target and inspection items from the data standard value storage unit 20 (FIG. 15) (step 57).
[0049]
FIG. 17 shows the configuration of the data standard value storage unit 20. The data standard value storage section 20 stores data on inspection targets, inspection items, and standard values. The standard value is a standard value of an inspection result for each inspection item, that is, a design standard value of equipment parts and the like. For example, the bolt size, which is one of the inspection items, changes only in a very limited range. In the case where the inspection result does not change much like this example, the standard value is defined in advance based on the design standard value and stored in the data standard value storage unit 20.
[0050]
In step 58 of FIG. 16, the difference between the standard value acquired from the data standard value storage unit 20 of FIG. 17 and the numerical value of the candidate word, which is the speech recognition result, is calculated. For candidate words whose difference is larger than a preset value, candidate ranks with lower differences are set in descending order of the difference. After changing the candidate order of the candidate words, the numerical data determination unit 17 outputs the candidate words and the candidate order to the inspection record management unit 15 in the next step 59.
[0051]
For example, for the inspection target “coupling” and the inspection item “bolt 1 dimension”, the candidate words “300.1” (first candidate), “200.1” (second candidate), and “200.0” ( It is assumed that the third candidate) is input to the numerical data determination unit 17. First, the numerical data determination unit 17 acquires a standard value corresponding to the inspection target “coupling” and the inspection item “bolt 1 dimension” from the data standard value storage unit 20. In the example shown in FIG. 17, the standard value is “200.00 mm”. Next, the difference between each of the candidate words “300.1”, “200.1”, “200.0” and the standard value “200.00” is calculated. At this time, if a preset value that the difference from the standard value is within 10% is provided in advance, the first candidate “300.1” does not satisfy this condition. The two candidate words whose difference from the standard value is within 10% are “200.1” and “200.0”. In this case, the setting of the candidate order of the candidate words is performed by voice recognition. According to the order of the candidate order determined by the unit 12. Therefore, the numerical data determination unit 17 sets “200.1” as a first candidate, “200.0” as a second candidate, and “300.1” as a third candidate.
[0052]
According to the present embodiment, when the inspection result of the numerical data is input by voice, the difference between the candidate word and the knowledge data on the configuration or function of the facility such as the standard value is determined, and the difference is larger than a preset value. Recognition rate can be improved by lowering the word candidate rank. As a result, it is possible to prevent the input numerical value from being recognized as an impossible numerical value and to improve the usability of the voice input system.
[0053]
Further, in the apparatus of the present embodiment, the apparatus can be easily configured only by adding an apparatus for evaluating the validity of the standard value of the inspection result and the numerical data to the conventional voice input system. Therefore, it is possible to easily develop the system.
[0054]
Furthermore, since the apparatus of the present embodiment has a function of calculating the difference between the input inspection result and the corresponding standard value, by adding a function of displaying this value to the user, the user can check the difference between the standard value and the standard value. The deviation can be confirmed, and the abnormality can be detected early. In the present embodiment, the comparison target of the candidate words is a standard value, but the recognition rate can be improved by the same processing even if the initial inspection result is used instead of the standard value. Further, in this case, since the deviation from the initial state can be confirmed, the progress of the abnormality can be confirmed with respect to the inspection result whose value changes over a long period of time.
[0055]
Although the present embodiment describes the case where the present invention is used for inspection of a plant, the present invention is not limited to this, and in a situation where numerical data having a certain standard value is input, the recognition rate of speech recognition is similarly increased. Can be improved. For example, the present invention can be applied to input of dimensions of parts used for assembling a device, input of a foreign exchange rate used in a financial institution, and the like.
[0056]
FIG. 18 is a configuration diagram of a speech data recognition device according to the fourth embodiment of the present invention. This embodiment is different from the first embodiment shown in FIG. 1 in that a priority word processing unit 21 that changes the candidate order of candidate words based on a check position of an input person and a word to be output with priority are stored. A priority word storage unit 22, a plant information storage unit 24 storing sensor installation positions and inspection routes, an inspection system information storage unit 25 storing words to be output with priority for each inspection target, and a plant information storage unit An inspection system identification unit 23 that identifies an inspection system using the unit 24 and outputs a word related to the inspection system acquired from the inspection system information storage unit 25 to the priority word storage unit 22 is provided. This is the point that the database 19 and the sensor installation information storage unit 18 are deleted.
[0057]
Hereinafter, the processing in the priority word processing unit 21, the configuration of the priority word storage unit 22, the configuration of the plant information storage unit 24, the configuration of the inspection system information storage unit 25, the processing of the inspection system identification unit 23, and the processing of the inspection record management unit 15 will be described. I do.
[0058]
FIG. 19 is a diagram showing sensor pre-stored data, which is data stored in advance in the plant information storage unit 24 shown in FIG. 18 and which is used for specifying the position of the voice input person. The sensor installation position data indicates which system each sensor is installed in. Incidentally, the sensors (S1 to S5) use a semiconductor laser and a photodetector, and are installed so that laser light always enters the light receiving surface of the photodetector. When the inspector (voice input person) passes between the semiconductor laser and the photodetector, the laser beam incident on the light receiving surface of the photodetector is blocked, so that the sensor passes the voice input person, that is, voice input. The position of the person.
[0059]
FIG. 20 is a configuration diagram of the inspection route information stored in the plant information storage unit 24 in advance. Inspection route information stores inspection routes in order by system name.
[0060]
FIG. 21 is a configuration diagram of the inspection system information storage unit 25. The inspection system information storage unit 25 includes an inspection system and priority words corresponding to the inspection system. The priority word is a word that is highly likely to be spoken by the inspector, and the priority word processing unit 21 outputs this word with priority. In the example of FIG. 21, “water supply piping”, “water supply temperature”, “water supply flow rate”, and the like are possible as priority words in water supply system inspection.
[0061]
FIG. 22 is a flowchart illustrating a processing procedure of the inspection system identification unit 23. The inspection system identification unit 23 operates when a signal from the sensor is detected (Step 60). First, a sensor that has detected a voice input person is specified, its set position information is read (steps 61 to 62), and inspection route information is read (step 63). Then, the current position of the voice input person, that is, the inspection system is specified (step 64). For example, in FIG. 23, the voice input person performing the inspection work along “Route 1” finishes the inspection of the circulating water pump A, and installs the sensor S2 to perform the inspection work of the T / D-RFP-A. When passing through the position, the sensor S2 makes a judgment based on the information "installed between the circulating water system and the water supply system" and the information on the route 1 "the water supply system after the circulating water system". The system is identified as “water supply system”.
[0062]
After specifying the system to be checked, a priority word corresponding to the system is acquired from the check system information storage unit 25 (step 65). The acquired priority word is stored in the priority word storage unit 22 (step 66). By this processing, words having a high possibility of utterance, that is, priority words, are sequentially stored in the priority word storage unit 22 in relation to the system under inspection.
[0063]
FIG. 24 is a flowchart illustrating a processing procedure of the priority word processing unit 21. The priority word processing section 21 operates when there is an input from the speech recognition section 12. The input data is candidate words and their candidate ranks. When the candidate word and the candidate ranking are input (step 70), the priority word processing unit 21 acquires a priority word from the priority word storage unit 22 (step 71). If the priority word is included in the candidate words, the candidate order of the word is set to the highest (step 72). However, when a plurality of priority words are included, the candidate order is assigned according to the candidate order determined by the speech recognition unit 12. After determining the candidate ranks of the candidate words in this way, the candidate words and the candidate ranks are output to the inspection record management unit 15 (step 73).
[0064]
For example, when the inspection system identification unit 23 specifies the inspection target as “water supply system”, the priority word storage unit 22 stores a word closely related to “water supply system”. These words are obtained from words stored in the inspection system information storage unit 25. In the example of the inspection system information storage unit 25 illustrated in FIG. 21, the priority words of “water supply system” are “water supply pipe”, “water supply temperature”, “water supply flow rate”, “water supply pump”, and the like. At this time, when “main steam flow” (first candidate), “water supply flow” (second candidate), and “water supply pump” (third candidate) are input to the priority word processing unit 21 as candidate words, The word processing unit 21 sets each of the candidate orders of “water supply flow rate” and “water supply pump” included in the priority words to be higher. However, since the voice recognition unit 12 sets the candidate order of “water supply flow rate” higher than “water supply pump”, the priority word processing unit 21 sets “water supply flow rate” as the first candidate and sets “water supply pump” to “water supply pump”. The second candidate is used. “Main steam flow rate” that is not included in the priority word is a third candidate. By performing the above processing, the candidate rank of the word related to the “water supply system” to be checked becomes higher, and the recognition rate is improved.
[0065]
FIG. 25 is a flowchart illustrating a processing procedure of the inspection record management unit 15. The inspection record management unit 15 operates when there is an input from the priority word processing unit 21 (step 75). The input data is a candidate word corresponding to a check target, a check item, a check result, and a candidate rank thereof. When an inspection target is input (step 76), an inspection record table corresponding to the inspection target is output to the input / output control unit 14 (step 81), and input of an inspection item is waited for (step 82). When an inspection item is input (step 77), the inspection item is output to the input / output control unit 14 (step 83), and input of an inspection result is waited for (step 84). The output of the inspection record table and the inspection items to the input / output control unit 14 is for displaying them on the portable display 29.
[0066]
When the inspection result is input (step 78), the inspection result is stored in the inspection record storage unit 16 (step 79). At the same time, the inspection result is output to the input / output control unit 14 (step 80). Here, the check result indicates a word having the highest candidate rank among candidate words. When the inspector inputs a word other than the inspection result by voice while the system is requesting the input of the inspection result, the inspection record management unit 15 outputs an error message displayed on the portable display 29 to the input / output control unit 14. (Step 85).
[0067]
According to the present embodiment, the position of the inspector is detected, the system targeted by the inspector is obtained from the position information, and only words related to this system are preferentially output, thereby improving the recognition rate. Becomes possible. Further, in the present embodiment, the words to be output are preferentially changed according to the dynamically changing situation of the position information of the inspector, so that the possibility of utterance is high according to the inspection object that changes every moment. By selecting a word, it is possible to reduce the possibility of misrecognition of the system, device name, and the like.
[0068]
Furthermore, in this embodiment, the first to third embodiments handle only numerical data, but it is possible to improve the recognition rate of speech recognition for words other than numerical data. With this, if one voice recognition dictionary that registers words related to the entire plant is prepared, the words that are most likely to be spoken are selected from similar system / device names according to the situation of the inspector. Can be output as a recognition word, the possibility of outputting a similar word as a recognition result can be reduced, and the man-machine nature of the voice input system can be improved.
[0069]
FIG. 26 is a configuration diagram of a speech data recognition device according to a fifth embodiment of the present invention. This embodiment is different from the fourth embodiment shown in FIG. 18 in that the inspection system identification unit 23 takes in the signal from the input / output control unit 14 and processes it, and the inspector replaces the plant information storage unit 24. The information storage unit 26 is provided. The system to be inspected is stored in the inspector information storage unit 26 for each inspector.
[0070]
In this embodiment, information for identifying the inspector, such as the inspector name or the inspector code, is input using the external input device 3. The system stores in advance the name of the system to be inspected for each inspector. By associating this information with the inspector, the system to be inspected is specified, and words related to the inspected system are identified. Output with priority.
[0071]
FIG. 27 is a configuration diagram of the inspector information storage unit 26. The inspector information storage unit 26 stores data of “inspection date”, “inspector”, and “inspection system”. For example, in the example of FIG. 27, the inspector A is to inspect the main steam system on April 10, 1995. This information is referred to when specifying the system to be inspected from the data of the “inspector”.
[0072]
FIG. 28 is a flowchart illustrating a processing procedure of the inspection system identification unit 23. The inspection system identification unit 23 operates when there is an input from the input / output control unit 14 (step 86). The input data is an inspector name input to the system using the external input device 3. When the inspector name is input, the inspection system identification unit 23 acquires a system corresponding to the inspector name from the inspector information storage unit 26 (Step 87). At this time, the inspection date corresponding to the current date is selected. Information on the current date and time is obtained from the clock in the inspection system identification unit 23. Next, a priority word corresponding to the system is acquired from the inspection system information storage unit 25 (step 88). Next, the acquired priority word is output to the priority word storage unit 22 (step 89).
[0073]
For example, when the inspector name “B” is input to the inspection system identification unit 23, the inspection system identification unit 23 refers to the information of the inspector name “B” in the inspector information storage unit 26. In the example shown in FIG. 27, the inspection system corresponding to the inspector name "B" is "water supply system". At this time, the inspection date that matches the current date is selected. Next, the inspection system identification unit 23 acquires the priority word of “water supply system” from the inspection system information storage unit 25 and stores the priority word in the priority word storage unit 22.
[0074]
FIG. 29 is a flowchart illustrating a processing procedure of the input / output control unit 14. When there is an input from the inspection record management unit 15 (step 90), the input / output control unit 14 outputs this signal to the transmission / reception unit (fixed station) 11 (step 95). This is for displaying the inspection record table or the words input by voice on the portable display 29. When a voice signal is input from the transmission / reception unit (fixed station) 11 (step 91), this signal is output to the voice recognition unit 12 in order to convert the voice signal into a character signal (step 96). When there is an input from the external input device 3 (step 92), if the input data is the inspector name (step 93), this is output to the inspection system identification unit 23 (step 97). If the input data is a request signal for outputting the inspection result to the external display device 4 or the printer 5, the inspection result is obtained from the inspection record management unit 15, and is output to the external display device 4 or the printer 5 ( Step 94). According to the present embodiment, the system to be inspected is specified based on the information related to the inspector, that is, the input code such as the type code and name of the voice input operator, and the words related to the inspection system are specified. Is preferentially output, so that the recognition rate of voice data can be improved.
[0075]
In the present embodiment, when the inspectors are grouped and the system for performing the inspection is determined by the groups, the information is stored for each group instead of storing the information for each inspector. By storing and inputting information for specifying a group such as a group name into the system, a system to be inspected can be specified in the same manner as in the above-described embodiment, and only words closely related to the system are prioritized. Can be output as desired. For example, if group "A" is a group that checks the main steam system, entering information identifying the group "A" into the system will cause the words associated with the main steam system to be entered. Can be output preferentially. Through these processes, the recognition rate of voice recognition can be improved.
[0076]
According to each of the above-described embodiments, when numerical data, system equipment names, and the like are input during plant operation, maintenance, and inspection work, not only the similarity to the standard pattern registered in the speech recognition dictionary, but also Information related to the characteristics of the object of the voice input system, such as the process value, the previous value of the inspection result, the standard value of the design, the position information of the voice input person, the information on the inspection target of the voice input user, etc. shall also be used. Accordingly, it is possible to improve the recognition rate of voice recognition according to the state of the plant that changes every moment, the inspection target, and the like. Thereby, the burden on the voice input person in the inspection work of the plant or the like, that is, the burden of re-input or the like is reduced, and the efficiency of these works can be improved. Further, the possibility of erroneous recognition of a voice input can be reduced, and a reliable check operation can be performed. Therefore, the effect of improving the economy and safety by using such a voice input system increases.
[0077]
【The invention's effect】
According to the present invention, when performing speech data recognition, the candidate ranking of the recognition result determined by the similarity with the standard pattern of the speech data is set to the characteristic such as the process amount targeted by the speech data to be recognized. Since the change is performed based on the speech data, the recognition rate of the voice data is improved.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a speech data recognition device according to a first embodiment of the present invention.
FIG. 2 is a flowchart showing a processing procedure in recording an inspection result.
FIG. 3 is a configuration diagram of an inspection record storage unit.
FIG. 4 is a schematic diagram illustrating a relationship between an inspection result and a process amount.
FIG. 5 is a flowchart illustrating a processing procedure in a numerical data determination unit.
FIG. 6 is a configuration diagram of a sensor installation information storage unit.
FIG. 7 is a configuration diagram of a plant database.
FIG. 8 is a flowchart illustrating a processing procedure in an inspection record management unit.
FIG. 9 is a flowchart illustrating a processing procedure in a voice recognition unit.
FIG. 10 is a flowchart illustrating a processing procedure in an input / output control unit.
FIG. 11 is a diagram illustrating a display example of a portable display.
FIG. 12 is a configuration diagram of a voice data recognition device according to a second embodiment of the present invention.
FIG. 13 is a configuration diagram of an inspection record storage unit according to a second embodiment.
FIG. 14 is a flowchart illustrating a processing procedure in a numerical data determination unit according to the second embodiment.
FIG. 15 is a configuration diagram of a voice data recognition device according to a third embodiment of the present invention.
FIG. 16 is a flowchart illustrating a processing procedure in a numerical data determination unit according to the third embodiment.
FIG. 17 is a configuration diagram of a data standard value storage unit according to the third embodiment.
FIG. 18 is a configuration diagram of a voice data recognition device according to a fourth embodiment of the present invention.
FIG. 19 is an explanatory diagram of sensor installation position information.
FIG. 20 is an explanatory diagram of inspection route information.
FIG. 21 is a configuration diagram of an inspection system information storage unit.
FIG. 22 is a flowchart illustrating a processing procedure in an inspection system identification unit.
FIG. 23 is a schematic view of a plant site.
FIG. 24 is a flowchart illustrating a processing procedure in a priority word processing unit according to a fourth embodiment.
FIG. 25 is a flowchart illustrating a processing procedure in an inspection record management unit according to the fourth embodiment.
FIG. 26 is a configuration diagram of a voice data recognition device according to a fifth embodiment of the present invention.
FIG. 27 is a configuration diagram of an inspector information storage unit according to a fifth embodiment.
FIG. 28 is a flowchart illustrating a processing procedure in an inspection system identification unit according to the fifth embodiment.
FIG. 29 is a flowchart illustrating a processing procedure in an input / output control unit according to a fifth embodiment.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Recognition apparatus main body, 2 ... Input / display apparatus, 3 ... External input apparatus, 4 ... External display apparatus, 5 ... Printer, 6 ... Control computer, 7 ... Plant, 11 ... Transmission / reception part (fixed station), 12 ... Voice recognition unit 13 Voice recognition dictionary 14 Input / output control unit 15 Inspection record management unit 16 Inspection record storage unit 17 Numeric data determination unit 19 Plant database 18 Sensor installation information storage unit , 20 ... data standard value storage unit, 21 ... priority word processing unit, 22 ... priority word storage unit, 23 ... inspection system identification unit, 24 ... plant information storage unit, 25 ... inspection system information storage unit, 26 ... inspector information Storage unit, 27: microphone, 28: transmitting / receiving unit (mobile station), 29: portable display.

Claims (12)

In a speech data recognition method for comparing an input voice pattern with a standard voice pattern for a word that has been input and determining a candidate word based on the similarity between the two, a candidate is obtained by comparing the input voice pattern with the standard voice pattern. When the word and the candidate rank of the candidate word are obtained, and the candidate word is numerical data, according to the relationship between the characteristic of the object related to the numerical data and the obtained candidate word, A speech data recognizing method characterized by performing a process of changing a candidate order of a candidate word and determining a candidate order of the candidate word. In a speech data recognition method for comparing an input speech pattern and a standard speech pattern with respect to a speech input word and determining a candidate word based on the similarity between the two, the present invention relates to a numerical range that can be taken by the input numeric data. A relational expression is predetermined, and when numerical data is input by voice and a plurality of candidate words for the input voice are determined by comparison with a standard voice pattern, the relational expression is satisfied from the plurality of candidate words. A method for recognizing speech data, wherein a candidate word is determined as a high-ranking candidate rank. A speech data recognition method that compares an input speech pattern with a standard speech pattern for a word that has been input and determines candidate words based on the similarity between the two. When a plurality of candidate words are determined by data input and comparison with a standard voice pattern, numerical data close to the previously recognized numerical data is determined as a higher candidate rank from among the plurality of candidate words. A voice data recognition method. In a speech data recognition method for comparing an input speech pattern and a standard speech pattern with respect to a speech input word and determining a candidate word based on the similarity between the two, measurement data for an object whose standard value is known in advance When a plurality of candidate words are determined by voice input and comparison with a standard voice pattern, a candidate word representing measurement data close to the standard value is determined as a higher candidate rank from among the plurality of candidate words. Recognition method of voice data as a feature. An object related to a task performed by a voice entrant in a voice data recognition method in which a voice input word is compared with an input voice pattern and a standard voice pattern and a candidate word is determined based on a similarity between the two . A word having a high possibility of utterance is determined in advance as a priority word based on , and when a plurality of candidate words are determined based on a speech recognition result using a standard voice pattern for an input voice , based on the object, A method for recognizing speech data, wherein a candidate word that matches a priority word is determined as a higher-rank candidate rank. In a speech data recognition method in which an input speech pattern is compared with a standard speech pattern and a candidate word is determined based on a similarity between the two, a plurality of candidate words are determined based on a speech recognition result. And determining a candidate rank of the plurality of candidate words based on characteristics of a speaker. In recognition device voice data comprising voice recognition means for determining candidate words on the basis of the similarity of the two compares the input speech pattern and the standard speech pattern to words that are voice input, a candidate word by the speech recognition means If the candidate word of the candidate word is numerical data when the candidate rank of the candidate word is obtained, the candidate word of the candidate word is determined according to the relationship between the characteristic of the object related to the numerical data and the obtained candidate word. An apparatus for recognizing speech data, comprising means for performing a process of changing a candidate rank and determining a candidate rank of the candidate word. In a voice data recognition device including a voice recognition unit that compares an input voice pattern with a standard voice pattern for a voice input word and determines a candidate word based on the similarity between the two, the input numerical data can be taken. A storage unit for storing in advance the data of the relational expression relating to the numerical range, and when the numerical data is input by voice and a plurality of candidate words for the input voice are determined by the voice recognition means, the relation is selected from among the plurality of candidate words. recognizer voice data, characterized in that it comprises means for determining a candidate word which satisfies the expression as a candidate ranking higher, the. In a speech data recognition device including speech recognition means for comparing an input speech pattern with a standard speech pattern for a speech-inputted word and determining a candidate word based on the similarity between the two, a speech input and a speech recognition performed in the past When numerical data for the same object is input as speech and a plurality of candidate words are determined by the voice recognition means, numerical data close to the previously recognized numerical data is determined as a higher candidate rank from among the plurality of candidate words. An apparatus for recognizing speech data, comprising: In a speech data recognition device including a speech recognition unit that compares an input speech pattern with a standard speech pattern for a word that has been input and determines a candidate word based on the similarity between the two, an object whose standard value is known in advance Means for deciding a candidate word representing measurement data close to the standard value from the plurality of candidate words as a higher candidate rank when a plurality of candidate words are determined by the speech data when measurement data for the object is input. An apparatus for recognizing voice data, comprising: In a voice data recognition device that includes a voice recognition unit that compares an input voice pattern with a standard voice pattern for a word that has been input and determines a candidate word based on the similarity between the two , A storage unit that stores in advance a term having a high possibility of speech based on a related object as a priority word, and the priority word, based on the object when a plurality of candidate words are determined by the voice recognition unit. recognizer voice data, characterized in that it comprises means for determining a matching candidate word candidate rank as a candidate ranking higher, the. In a speech data recognition apparatus comprising: a speech recognition unit that compares an input speech pattern with a standard speech pattern for a word that has been input and determines a candidate word based on a similarity between the two. A speech data recognition apparatus comprising: means for determining a candidate rank of the plurality of candidate words based on characteristics of a speaker when determined by the means.

Images