JP5049930B2 - Distributed speech recognition system - Google Patents

Description

  The present invention relates to a technique for giving a voice command to a plurality of control devices that mainly control operation of a plant such as power generation.

  A control device (also referred to as a “control panel”) for controlling the operation of the plant is arranged at a site where the plant is arranged, and usually a plurality of control devices are arranged in parallel. In this control device, signals are exchanged with other control devices when performing the control.

  In recent years, signal communication between control devices can be switched from wired communication in which control devices are connected by cable wiring to wireless communication. In the case of wireless communication, it is not necessary to provide an interface or space necessary for cable wiring. Basically, it is sufficient to arrange a transmission device and a reception device with respect to the control device. However, in wireless communication, it is necessary to unify the standards of control devices and the like. When control devices with different standards are used from different manufacturers, it is difficult to realize a system composed of these devices.

  Therefore, a method of using voice in wireless communication can be considered. In other words, a command for the control device is output to the control device by voice from a speaker arranged at the site, and signal exchange between the control devices including voice signal processing is realized. With this method, the transmitting device is not necessary, and only the receiving device needs to be supported, and it is easy to standardize the standard in realizing the system.

  Such a system is particularly useful in inspection work for control devices. In this inspection work, a specific inspector brings test equipment to the site, confirms the operation of the joint, and confirms the operation of auxiliary equipment that is the control target of the control device, that is, pumps, valves, etc. constituting the plant. The control device is inspected. In this inspection, an inspector at the site and a monitor who is monitoring at a factory away from the site communicate with each other through a transceiver or the like. The inspector performs the inspection in accordance with the instruction from the observer and reports the result to the observer. In some cases, the supervisor may perform a command operation for directing the control device to support inspection by the inspector. If this command operation is performed by voice, the inspection work becomes easy.

  However, when voice is used, it is necessary to properly recognize the voice on the control device side, but noise enters the voice and may cause erroneous recognition of the voice. This is especially true at sites where a lot of noise is generated by other operations.

As disclosed in Patent Document 1, if speech is detected at a plurality of locations and recognized while comparing the speech, erroneous recognition can be reduced to some extent. However, since the configuration for processing the speech recognition for the detection of the speech is simplified, the recognition accuracy required for the inspection is realized unless a highly accurate speech recognition technology is used. It is difficult.
Japanese Patent No. 3725566 (paragraph 0025, FIG. 4 etc.)

  In view of the above circumstances, an object of the present invention is to improve the accuracy of recognition of voice input to a plant control device such as power generation.

  In order to achieve the above object, according to the present invention, when speech recognition is performed in a plant control device such as power generation, it is determined whether or not the input speech signals coincide with each other. Details will be described later.

  According to the present invention, it is possible to improve the accuracy of recognition of a voice input to a plant control device such as power generation.

≪Configuration≫
Embodiments of the present invention (hereinafter referred to as “embodiments”) will be described below with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of a network in which the distributed speech recognition system according to the present embodiment is used.

  This distributed speech recognition system controls the operation of a plant such as power generation, and is mounted on a plurality of control devices 2 arranged in parallel (eight in FIG. 1) in the field. The distributed speech recognition system will be described later. In addition to the control device 2, the on-site network includes a voice output unit 1 that outputs at least voice output from the factory to the control device 2, a computer (such as a factory server 8) in the factory, and an Internet line. The on-site server 4 connected to be communicable is connected to be communicable. In addition, the inspector 3 is inspecting the control device 2 at the site. As an inspection, for example, an AI / O (Analog Input / Output) unit or DI / O (Digital Input / Output) unit implemented as a PI / O (Process Input / Output) interface mounted on the control device 2 is used. There is a static characteristic inspection.

  On the other hand, in a factory, a factory server 6 having a microphone 7 for inputting voice and a factory computer 6 (site server 4 etc.) connected to the network so as to be communicable via an Internet line. Are communicably connected. In the factory, a supervisor 5 who controls the operation status of the control device 2 by voice is on standby. In addition, the supervisor 5 and the inspector 3 communicate with each other mainly regarding the inspection of the control device 2 using a transceiver, for example.

  In addition, the audio | voice output part 1, the control apparatus 2, the site server 4, the factory terminal 6, and the factory server 8 which are computers are an input part which has an input port, an output part which has an output port, as the hardware constitutions, A control unit (including a first control unit and a second control unit) realized as a CPU (Central Processing Unit) and the like, and a storage unit (first storage) realized as an HDD (Hard Disk Drive) as an external storage device 1 and a second storage unit), and a memory realized by a RAM (Random Access Memory) having a storage area where data to be read and written is expanded. When processing relating to the present invention is executed in these devices, a program stored in the storage unit of each device is loaded into the memory and executed by each CPU (control unit), so that each device constituting the network It is executed by each processing unit embodied in Each program may be stored in the storage unit in advance, or may be introduced when necessary via another storage medium or communication medium (a network or a carrier wave propagating through the network).

  FIG. 2 is a block diagram showing an example of the configuration of the distributed speech recognition system according to the present embodiment. Each of the control devices 2 includes a speech input unit (A1 or B1), a speech recognition unit (A2 or B2), a time application unit (A3 or B3), and a speech normality determination unit as functional units constituting the distributed speech recognition system. (A4 or B4), a message processing unit (A5 or B5), an important voice table (A6 or B6), a recognized voice table (A7 or B7), and a command message table (A8 or B8). The two control devices 2 (first control device and second control device) shown in FIG. 2 are connected to the control device 2 (shown in FIG. 1) to which the sound output from the sound output unit 1 is input. 8 control devices) are selected by a predetermined method. As a selection method, for example, there is a method in which the volume of the sound input to each control device 2 is measured and the control device 2 to which the highest volume is input is selected. In this case, a device for measuring the volume is provided in each control device 2.

  The voice input unit (A1, B1) has a function of detecting the voice of the supervisor 5 output from the voice output unit 1 including noise and other sounds. The detected voice is converted into, for example, a digital voice signal, and the voice signal is output to the voice recognition unit (A2, B2).

The voice recognition unit (A2, B2) performs frequency analysis on the voice signal output from the voice input unit (A1, B1) using, for example, FFT (Fast Fourier Transform) or the like, and extracts voice feature values. To do. If the feature quantity cannot be extracted, it is recognized as a voice consisting only of noise, and the voice signal is ignored. Ignored audio signals are discarded. Speech recognition is performed by using, for example, a dictionary (not shown) stored in each control device 2 by a technique such as a composite similarity method, HMM (Hidden Markov Model), DP (Dynamic Programming) matching using the feature amount. Keywords are systematically managed and collated with a stored data group).
In addition, in the voice signal from which the feature amount is extracted, if there is a voice that matches the voice registered in the important voice table (A6, B6) to be described later, the voice is processed as being included in the voice, The audio signal is output to the time application unit (A3, B3). The important voice will be described later.

  The time application unit (A3, B3) adds time data indicating the time when the voice is input to the control device 2 to the voice signal output from the voice recognition unit (A2, B2). The time data is data generated when a timer (not shown) installed in each control device 2 performs a time measurement process. For example, when the voice input unit (A1, B1) requests the timer to generate time data, a time measurement process is performed, and control is performed such that the time when the request is received is generated. The voice signal to which the time data is added is temporarily registered in the recognized voice table (A7, B7) with a predetermined data structure.

  The sound normality determination unit (A4, B4) determines whether or not the sound signal acquired from the recognized sound table (A7, B7) is a normal signal in which noise mixing is a predetermined threshold value or less. In this determination, the audio signal is received from the audio normality determination unit of the other control device 2, and the time indicated by the time data added to the audio signal received by itself is substantially the same as that received from the other control device. For example, audio signals at times falling within ± 0.2 seconds are compared. The audio signal determined to be normal by this comparison is output to the message processing unit (A5, B5), and the other is ignored. In the comparison, for example, it is determined whether or not the rate at which the waveforms of two audio signals to be compared match is equal to or greater than a threshold value.

  The message processing unit (A5, B5) controls the operation of the auxiliary machine 9, for example, the audio signal output from the audio normality determination unit (A4, B4) is used for processing executed by the control device 2. It is determined whether or not the audio signal is used for the recording. This determination is made as to whether or not the audio signal includes the command content (described later) registered in the command message table (A8, B8). If registered, data indicating the contents of the command (control data) is output to the auxiliary machine 9 to control the auxiliary machine 9, otherwise the audio signal is ignored.

The important voice table (A6 or B6) registers the voice of the voice signal determined to be important for controlling the auxiliary machine 9 as the important voice and has a predetermined data structure (for example, converted into a voice file). It is a database. The determination of whether or not it is important is a design matter of the person who controls the plant, but usually the voice that defines the operation of the auxiliary machine 9 is important. For example, it refers to “starting” and “stopping” of the auxiliary machine 9.
FIG. 3 shows the data structure of the important voice table (A6, B6). The important voice table (A6 or B6) is a database including an identification number (No) field 301 in which a number for identifying the record is registered, and an important voice field 302 in which a voice defined as important is registered. is there. If there is a portion that matches the sound registered in the important sound field 302 in the sound of the input sound signal, flag processing is performed to determine that portion as the important sound.

The recognized voice table (A7, B7) is a database that registers a voice signal to which time data is added, having a predetermined data structure (for example, associating a voice signal that has been converted into a voice file and time data). is there.
FIG. 4 shows the data structure of the recognition voice table (A7, B7). The recognition voice table (A7, B7) includes an identification number (No) field 401 in which a number for identifying the record is registered, a time field 402 in which the time indicated by the time data is registered, and the voice indicated by the voice signal. The database includes an audio field 403 in which a part that is not important voice is registered, and an important voice field 404 in which a part that is important voice among voices indicated by the voice signal is registered. After the auxiliary machine 9 is controlled for the voice signal, the record in the recognized voice table (A7, B7) is discarded.
Note that the voice normality determination unit (A4, B4) is a case where it is determined that the audio signal is disturbed by a slight amount (for example, about 30%) of noise in the non-important voice part. Even if it is judged as normal. However, with respect to a portion that is an important voice, a difference is provided in recognition accuracy so that it is judged normal when the noise mixed in the voice signal is a very small amount (for example, about 1%) or less.

In the command message table (A8, B8), each control device 2 registers the command contents to be commanded (mainly, the auxiliary machine 9) having a predetermined data structure (for example, converted into a voice file). It is a database.
FIG. 5 shows the data structure of the command message table (A8, B8). This command message table (A8, B8) includes an identification number (No) field 501 in which a number for identifying the record is registered, and a target facility number field in which a value for identifying a facility to be commanded, that is, a plant is registered. 502, a command target field 503 in which a value for identifying the auxiliary machine 9 constituting the plant is registered as a command target of the control device 2, and a command field 504 in which a value for identifying the command content from the control device 2 is registered. And a logic sheet field 505 in which a logic sheet (algorithm) expressing the command content in a logic circuit is registered, and a number for identifying a voice command input to the logic sheet in order to realize control according to the command content The database includes a voice command number field 506 to be registered. The number registered in the voice command number field 506 has a one-to-one correspondence with the value registered in the command field 504.
The logic sheet is stored in the storage unit of each control device 2, and when a record that matches the input audio signal is extracted, the logic sheet registered in the record is read out. Further, the logic sheet has an area as a logic circuit in which the number registered in the voice command number field 506 is input.

<< Process >>
Next, processing operations performed by the distributed speech recognition system according to the present embodiment will be described. FIG. 6 is a flowchart showing the processing operation by the distributed speech recognition system according to the present embodiment. The main body of this processing operation is the control unit of each control device 2.
In addition, this processing operation is performed by the inspection device 3 by the inspector 3 at the site and the monitoring device 5 monitoring at the factory (mainly, control of the control device 2 with respect to the auxiliary machine 9 (including interaction)). The description will be made assuming that the test is performed in the inspection for confirming whether or not the test is normally performed. By the way, the inspector 3 and the monitor 5 communicate with each other about the inspection through the transceiver. For example, the inspector 3 gives the
"Board No.5 Terminal block No.X11 Jumper No.1 and No.2"
And,
“Panel No.5 Terminal block No.X11 Channel No.1 2.5V input”
In order to check the control device 2, the work content for confirming a predetermined signal input is reported.
On the other hand, the supervisor 5 changes the operating state of the auxiliary machine 9 according to the work of the inspector 3 described above.
"No. 1 water supply pump A starts"
A command necessary for the test of the auxiliary machine 9 is output from the voice output unit 1 by voice. This sound is input to all the control devices 2 although the volume varies.

  First, in step S01, the control unit of the control device 2 detects the voice of the supervisor 5 transmitted from the factory via a network or the like and output from the on-site voice output unit 1, including noise and other sounds. To do. The detected voice is input as a voice signal. After the input, the process proceeds to step S02.

  Next, in step S02, the control unit of the control device 2 determines whether or not the input audio signal can be recorded in the memory. If recording is possible (Yes in S02), the process proceeds to step S03. Otherwise (No in S02), the process proceeds to step S09. Note that the voice signal can be recorded means that the voice recognition unit (A2, B2) can perform frequency analysis and extract a voice feature amount. In addition, referring to the important sound table (A6, B6), if the sound signal includes a feature amount of sound corresponding to the important sound such as “Kidou”, the flag of the important sound included in the sound signal. Process. For example, when a voice saying “No. 1 water supply pump A starts” is input, it is analyzed by voice recognition as “Strawberry Equip Pump Ikido”, and “Kido” is the important voice table (A6, B6). ) No. 1 record (see FIG. 3) is referred to and flag processing is performed as an important signal. Note that the part of “I do” is ignored (see step S09).

  Next, in step S03, the control unit of the control device 2 applies time to the audio signal determined to be recordable, and generates an audio signal with time data. The voice signal with time data is temporarily registered in the recognized voice table (A7, B7). After applying the time, the process proceeds to step S04. When the voice “Start No. 1 water supply pump A” is input at 10:10, a record (see FIG. 4) numbered No. 1 in the recognition voice table (A7, B7) is created.

  Next, in step S <b> 04, the control unit of the control device 2 compares the sound signal near the time input by the other control device 2. In this comparison, an audio signal with time data is acquired from another control device 2, and the audio that can be determined to be substantially the same time by referring to the time registered in the time field 402 of the recognized audio table (A7, B7). Extract the signal and compare the two. After the comparison, the process proceeds to step S05.

  Next, in step S05, the control unit of the control device 2 determines whether or not the audio signals to be compared match. If the audio signals match (Yes in step S05), it is determined as a normal audio signal, and the process proceeds to step S06. Otherwise (No in step S05), it is determined that the sound signal is abnormal, and the process proceeds to step S10. Note that the voice signals match each other, the voice (voice other than the important voice) included in the voice signal with time data acquired from the other control device 2 and the voice field of the corresponding record in the recognized voice table (A7, B7). The audio registered in 403 matches even if some noise is included (the ratio of the audio signals to be compared is equal to or higher than the second threshold but lower than the first threshold), and other The important voices included in the voice signal with time data acquired from the control device 2 and the important voices registered in the important voice field 404 of the corresponding record of the recognition voice table (A7, B7) contain almost no noise and are almost strictly (The ratio at which the audio signals to be compared are equal to or higher than the first threshold value). When the voice “No. 1 water pump A starts” is input, it will be recognized as “Kyu Sui Pump” on the one hand, and as “Cous XX Pump” (XX part is noise) on the other hand. However, since it is not an important voice, it is determined that they match. However, if it is recognized as “Kido” on the one hand and “× Doe” on the other hand (X is a noise), it is determined that the voices are not matched because they are important voices.

  Next, in step S06, the control unit of the control device 2 compares the audio signal determined to be normal with the data in the command message table (A8, B8). The command message table (A8, B8) is searched using the feature value of the voice included in the voice signal determined to be normal as a search key. After the comparison, the process proceeds to step S07.

Next, in step S07, the control unit of the control device 2 determines whether or not the target record exists in the command message table (A8, B8). Specifically, the voice feature amount included in the voice signal serving as the search key matches the values registered in the target equipment number field 502, the command target field 503, and the command field 504 in the command message table (A8, B8). Judge whether to do. If the record exists (Yes in step S07), the process proceeds to step S08. Otherwise (No in step S07), the process proceeds to step S11.
In the present embodiment, it is assumed that the value registered in the command target field 503 is a voice that is not an important voice, and the value registered in the command field 504 is an important voice. When a voice saying “No. 1 water supply pump A starts” is input, the part of “Strawberry cucumber pump A” matches the value of the command target field 503 of the record numbered No1, and “Kido” The part matches the value of the command field 504.

  Next, in step S08, the control unit of the control device 2 outputs the data described in the command message table (A8, B8) to the auxiliary machine 9. Specifically, in the record extracted from the command message table (A8, B8) by the voice signal serving as the search key, the logic sheet registered in the logic sheet field 505 is read from the storage unit and stored in the voice command number field 506. The registered voice command number is input to a predetermined area of the read logic sheet. If input, control data that is an operation result is obtained in accordance with the logic circuit shown in the logic sheet. The control data is output to the auxiliary machine 9 and the process ends. When a voice saying “No. 1 water pump A starts” is input, a record numbered No1 is referred to, a logic sheet “CS001” and a voice command number “1” are extracted, and a logic sheet CS001 is extracted. The voice command number 1 is input to the predetermined area. As a result, control data for starting the feed pump A of the plant 1 is output from the control device 2.

  Next, in step S09, the control unit of the control device 2 ignores the audio signal determined to be unrecordable, discards the audio signal, and ends.

  Next, in step S <b> 10, the control unit of the control device 2 ignores the audio signal determined not to match the other control device 2, discards the audio signal, and ends.

Next, in step S11, the control unit of the control device 2 ignores the audio signal determined to have no target record in the command message table (A8, B8) for the input audio signal, The audio signal is discarded and the process ends.
This is the end of the description of the processing operation performed by the distributed speech recognition system.

≪Summary≫
According to the present embodiment, the following effects can be obtained. That is, when performing speech recognition in a plant control device such as power generation, it is determined whether or not the voice signals of the input speech match between the control devices, so that they are input to the plant control device such as power generation. The accuracy of voice recognition can be increased. When raising the accuracy of speech recognition, the idea of how to recognize accurately with the system “single unit” is the mainstream, but when used for plant control devices such as power generation, the difficulties described above arise, A desired recognition accuracy is achieved by providing “plural” speech recognition locations. Since the hardware configuration originally includes a plurality of control devices, the burden of reducing resources, adding interfaces, and the like due to the provision of a plurality of distributed speech recognition systems is small.

  Further, by controlling the control device by voice recognition, it is not necessary to use cable wiring to connect to another control device, and the application range of the distributed speech recognition system of this embodiment is expanded.

  In addition, since the communication between the control devices is performed by voice, there is no limit on the number of signals to be connected, and no wiring is required, so there is no need to examine the current value of the current flowing through the wiring or to perform a connection test. At the same time, the entire apparatus can be greatly reduced in size.

≪Others≫
In addition, although the said form is the best thing for implementing this invention, the implementation form is not limited to this. Therefore, various modifications can be made to the implementation form without changing the gist of the present invention.

  For example, in the present embodiment, the voice output unit 1 outputs a voice commanded to the control device 2 by the monitor 5 using the microphone 7. However, this sound may be recorded in advance for the command at the factory terminal 6, read out at a desired timing, and output from the sound output unit 1.

  In the present embodiment, two control devices are selected from two or more control devices arranged in parallel, and voice recognition is performed in each control device. However, the number of control devices that perform speech recognition is not limited as long as it is plural. It is also possible to determine whether or not the voice signal is normal by performing voice recognition in all the control devices while considering the load associated with the processing. Further, the distributed speech recognition system according to the present embodiment is not mounted on the control device, but is configured as a plurality of independent devices, and a network in which these devices are communicably connected to the control device is configured. Also good.

  In addition, specific configurations such as hardware, software, flowcharts, and data structures can be changed as appropriate without departing from the spirit of the present invention.

  The distributed speech recognition system of the present invention can be applied to a device that executes processing by speech. When applied, it does not matter whether it is externally connected to the device or mounted inside the device.

It is the block diagram which showed the structure of the network where the distributed type speech recognition system by this embodiment is used. It is the block diagram which showed one Example of the structure of the distributed speech recognition system by this embodiment. The data structure of the important voice table (A6, B6) is illustrated. The data structure of a recognition voice table (A7, B7) is illustrated. The data structure of a command message table (A8, B8) is illustrated. It is a flowchart which shows the processing operation by the distributed speech recognition system by this embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Audio | voice output part 2 Control apparatus (A 1st control apparatus and a 2nd control apparatus are included.)
3 Inspector 4 On-site server 5 Monitor 6 Factory terminal 7 Microphone 8 Factory server 9 Auxiliary machine A1, B1 Voice input part A2, B2 Voice recognition part A3, B3 Time application part A4, B4 Voice normality judgment part A5, B5 Message processor A6, B6 Important voice table A7, B7 Recognition voice table A8, B8 Command message table

Claims (3)

A distributed speech recognition system comprising: a first control device that controls a first device; and a second control device that controls a second device,
The first control device includes:
The voice inputted from the outside is converted into a first voice signal, the voice is recognized by extracting the feature amount of the voice from the first voice signal, and the recognized voice receives a predetermined command. including satisfies the command containing conditions are in the first audio signal meets the normal condition is normal, and when it is determined to be within the first information identifying the first device in the audio signal Includes a first control unit that controls to output control data according to the command to the first device;
A first storage unit that stores a command for the first device and an algorithm for obtaining the control data in association with each other;
Equipped with a,
The second control device includes:
A voice inputted from outside is converted into a second voice signal, a voice feature is extracted from the second voice signal, the voice is recognized, and the recognized voice receives a predetermined command. including satisfies the command containing conditions are in, the second audio signal meets the normal condition is normal, and when it is determined that includes the information for specifying the second device to a second audio signal Includes a second control unit that controls to output control data in accordance with the command to the second device;
A second storage unit that stores a command for the second device and an algorithm for obtaining the control data in association with each other;
Equipped with a,
Before SL first controller and the second controller is communicatively connected,
When the first control unit determines that the command inclusion condition is satisfied, the first control unit adds first time data indicating a time when the sound is input to the first sound signal. ,
When the second control unit determines that the command inclusion condition is satisfied, the second control unit adds second time data indicating a time when the sound is input to the second sound signal. ,
The first controller is
Of the first audio signal to which the first time data is added and the second audio signal to which the second time data is added and acquired from the second control device, the second audio signal Comparing whether the time indicated by the first time data substantially matches the time indicated by the first time data to determine whether or not the normal condition is satisfied for the first audio signal. Distributed speech recognition system.
The first controller is
The first important audio signal included in the first audio signal by the comparison and corresponding to the command matches the second important audio signal included in the second audio signal and corresponding to the command. der ratio than the first threshold value is, and,
According to the comparison, the proportion of the first non-important audio signal included in the first audio signal and the second non-important audio signal included in the second audio signal is equal to the first threshold value. 2. The distributed speech recognition system according to claim 1, wherein the normal condition is determined to be satisfied for the first speech signal when the second threshold is less than or equal to a second threshold value.   There are a plurality of control devices that are candidates for the second control device,
  Each of the control devices includes volume measuring means for measuring the volume of the sound input to itself,
  Of each of the sound volume measuring means, a control device provided with sound volume measuring means for measuring the maximum sound volume is defined as the second control device.
  The distributed speech recognition system according to claim 1 or 2, wherein the system is a distributed speech recognition system.

Images