JP2017084041A - Voice command input device and welding system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To highly accurately input a voice command without being influenced by noise.SOLUTION: A voice command input device comprises: an acceleration sensor 101 that is fitted around the mouth of a speaker; a coordinate transformation part 102 for transforming a detection value of the acceleration sensor 101 into data of motions of the mouth of the speaker; a voice conversion part 104 for converting the data of the motions of the mouth into a string of phonemes by a first conversion table that is stored in a conversion table storage part 103; a voice recognition part 105 for recognizing a command word from data of the string of phonemes by a language model that is stored in a language model storage part 106; and a command signal conversion part 107 by which the recognized voice command is converted into a command signal and outputted while using a second conversion table that is stored in a command word storage part 108. The voice command of the speaker is recognized from motions or shapes of the mouth, such that a command signal corresponding to the voice command can be highly accurately inputted without being influenced by noise.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a voice command input device using an acceleration sensor and a welding system using the voice command input device.

  Conventionally, a welding system capable of inputting a voice command (hereinafter referred to as a “voice command”) has been proposed. For example, Patent Document 1 provides an ear-hook type microphone for an operator to input a voice command, while providing a microphone connection portion for connecting the microphone to a welding torch, and the operator performs a welding operation. A configuration is disclosed in which a microphone is connected to a microphone connection portion of a welding torch.

  Patent Document 1 transmits a voice command input to a microphone by an operator to a control circuit in the welding power supply device via a microphone connection portion and a connection cable that connects the welding torch to the welding power supply device. In this configuration, voice command input processing is performed.

  The welding site is an environment in which relatively high noise is generated, including noise caused by arc generation. For this reason, in order to control the welding power supply device of the welding system by a voice command (voice command mixed with noise) issued by an operator, noise based on noise is sufficient for the command signal obtained by converting the voice command into an electrical signal. It is necessary to suppress it.

  Conventionally, in a speech recognition system, a technique for reducing misrecognition and malfunction due to environmental noise has been proposed. For example, in Patent Document 2, as an effective method for reducing misrecognition and malfunction due to environmental noise, a voice microphone and a noise microphone are provided, and noise obtained with a noise microphone from voice obtained with the voice microphone is disclosed. A method of subtracting a component (noise component) (noise canceller) and a method of picking up only a user's voice using a microphone having a high directivity or a bone conduction microphone are disclosed.

  Patent Document 3 discloses a technique for recognizing a voice produced by a person using the movement of a lip of a person's photographed image. Specifically, in self-timer shooting in which a photographer himself / herself is photographed by a photographing apparatus, a lip region image is extracted from a moving image of the photographer as a subject, and the photographer selects a keyword related to the shutter from the movement of the lips of the extracted image. A technique for detecting the occurrence of an image and capturing a still image is described.

JP 2013-248663 A Japanese Patent Laid-Open No. 04-276799 Japanese Unexamined Patent Publication No. 2011-14985

  For example, in pulse welding, since the welding current is alternately switched between a small current and a large current at the pulse frequency, the thickness of the arc changes according to the change in the welding current. For this reason, the air around the arc vibrates and a very large noise is generated. The magnitude of this noise may be as high as 80 to 100 [dB] in some cases, and noise and noise larger than general factory noise are generated in pulse welding.

  Since very large noise is generated at the welding site, it is difficult to sufficiently remove the noise by performing filtering processing on the audio signal collected by the microphone. For this reason, in the voice command input device applied to the welding system, it is difficult to input a voice operation command with high accuracy.

  As a method not affected by noise, it is conceivable to apply a method disclosed in Patent Document 3 that reads a voice generated by a worker using a moving image obtained by photographing the worker's face.

  However, in this method, it is necessary to arrange the camera so as to face the operator's face, and it is not easy to arrange the camera. In particular, since the posture of the worker always changes during welding work, it may happen that the face part of the worker cannot be photographed with a camera fixed at a position different from the worker, and the voice produced by the worker cannot be read. Occurs.

  In order to avoid this inconvenience, when the camera is attached to the operator's head, the camera and the camera support member are welded to the operator because the camera is attached to face the operator's face. The problem of getting in the way of work arises.

  The present invention has been made in view of the above problems, and uses a voice command input device and a voice command input device that can input voice commands with high accuracy without being affected by noise around the work site. The purpose is to provide a welding system.

  A voice command input device according to the present invention is a voice command input device for inputting a command signal by voice, and a detection means for detecting a movement of the speaker's mouth when the speaker utters a voice command; Command word conversion means for converting the mouth movement detected by the means into a command word uttered by the speaker, and command signal generation means for generating a command signal corresponding to the command word converted by the conversion means. Features.

  According to this configuration, by recognizing the voice command uttered by the speaker from the movement of the speaker's mouth, it is possible to input a command signal corresponding to the voice command without being affected by ambient noise.

  In the above voice command input device, the detection means includes an acceleration sensor attached to the speaker's cheek or peripheral portion of the mouth, and a detected acceleration value output from the acceleration sensor in response to the speaker speaking the voice command. The first conversion means for converting the voice into the mouth movement data of the speaker, the command word conversion means is a first indicating a correspondence relationship between the phoneme uttered by the speaker and the mouth movement corresponding to the phoneme. Output from the first conversion means using the first storage means for storing the first conversion table, the second storage means for storing one or more command words registered in advance, and the first conversion table. A second conversion means for converting the speech movement data of the speaker into a phoneme string, and collating the phoneme string converted by the second conversion means with a command word registered in the second storage means. Voice recognition means for recognizing command words spoken by a speaker, The command signal generating means includes a third storage means for storing a second conversion table indicating a correspondence relationship between one or more command words stored in the second storage means and the command signal, and a second conversion table. And a third conversion means for converting the command word recognized by the voice recognition means into a command signal corresponding to the command word.

  According to this configuration, the phoneme constituting the voice command uttered by the speaker can be appropriately detected and detected from the data of the mouth movement of the speaker detected by the acceleration sensor and the first conversion table. The command word of the voice command can be appropriately recognized from the constituent phonemes of the voice command and the registered command word. Further, the recognized command word can be converted into a command signal corresponding to the command word and input using the second conversion table.

  Furthermore, in the above voice command input device, the voice recognition means calculates the degree of coincidence between the command word registered in the second storage means and the phoneme string converted by the second conversion means, and has a predetermined threshold value. The command word having the above matching degree and the highest matching degree may be recognized as a voice command uttered by the speaker.

  According to this configuration, even if a command word that completely matches the command word recognized by voice recognition is not registered, the voice command can be converted into a command word corresponding to the voice command and input with high accuracy. it can.

  In the voice command input device, the mouth movement data corresponding to the phoneme uttered by the speaker is an acceleration sensor based on a change in facial expression when the speaker utters the speech based on an expression that does not utter the speech. The displacement vector data may be good.

  According to this configuration, the movement of the mouth according to the phoneme of the voice command uttered by the speaker can be detected with high accuracy.

  In the voice command input device described above, the acceleration sensor is a triaxial acceleration sensor that detects acceleration in three axial directions orthogonal to each other, and may be attached to both sides of the speaker's cheeks or mouth.

  According to this configuration, it is possible to detect a delicate movement of the mouth corresponding to the phoneme of the voice command uttered by the speaker with high accuracy.

  A welding system according to the present invention supplies a welding torch, a wire feeding device that feeds a welding wire to the welding torch, and electric power for welding between the welding torch and the base material via the wire feeding device. A welding system comprising a welding power source and a command input device for inputting various commands to the welding power source, wherein the command input device comprises the voice command input device according to the present invention described above. It is characterized by.

  According to this configuration, by recognizing the voice command uttered by the speaker from the movement of the worker's mouth, the command corresponding to the voice command can be given to the voice command related to welding with high accuracy without being affected by ambient noise. A welding system that can convert a signal and input it to a welding power source can be applied.

  In the above welding system, the voice command input device may be provided in the wire feeding device.

  According to this configuration, the voice command input device is easily and compactly arranged in the welding system by providing the voice command input device in the wire feeding device arranged near the worker.

  The welding system may include a remote operation device that can be connected to the wire feeding device by wireless communication, and the voice command input device may be provided in the remote operation device.

  According to this configuration, the voice command input device is easily and compactly arranged in the welding system by providing the voice command input device in the remote control device held by the operator.

  According to the voice command input device and the welding system of the present invention, a voice command issued by an operator is recognized with high accuracy without being affected by noise, and a highly reliable command signal is generated based on the recognized voice command. Can be generated.

The block diagram which shows the structure of the voice command input device which concerns on this invention External view showing the external appearance of the acceleration sensor used in the voice command input device The figure which shows the detection direction of an acceleration at the time of attaching the acceleration sensor used for the voice command input device to a speaker's face The figure which shows the displacement vector of the acceleration sensor used for the voice command input device Figure showing an example of the shape of the mouth when the speaker utters a vowel The figure which shows an example of the 1st conversion table which shows the correspondence of the phoneme memorize | stored in the same voice command input device and the data of the movement of the mouth The block diagram which shows the structure of a welding system provided with the voice command input device The figure which shows an example of the data of the term used for the command word memorize | stored in the voice command input device, and the word The figure which shows an example of the command word of the welding system registered into the voice command input device, and the content Flowchart showing voice command input control by the voice command input device The figure which shows the modification of the welding system provided with the voice command input device

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of a voice command input device according to the present invention and a welding system including the voice command input device will be described with reference to the drawings.

  First, the voice command input device will be described.

  FIG. 1 is a block diagram showing a configuration of a voice command input device according to the present invention. FIG. 2 is an external view showing an external appearance of an acceleration sensor used in the voice command input device.

  The voice command input device 1 includes an acceleration sensor 101, a coordinate conversion unit 102, a conversion table storage unit 103, a voice conversion unit 104, a voice recognition unit 105, a language model storage unit 106, a command signal conversion unit 107, and a command word storage unit 108. Including.

  The acceleration sensor 101 is attached to the cheek of the speaker (worker) Q who issues a voice command and the periphery of the mouth, and detects movements of the cheek muscle that moves the corner of the mouth and the mouth muscle that creates the facial expression of the mouth. It is. The coordinate conversion unit 102, the conversion table storage unit 103, the voice conversion unit 104, the voice recognition unit 105, the language model storage unit 106, the command signal conversion unit 107, and the command word storage unit 108 are included in the device main body 100 of the voice command input device 1. Is provided.

  The acceleration sensor 101 and the coordinate conversion unit 102 correspond to a “detection unit” according to the present invention, and the coordinate conversion unit 102 corresponds to a “first conversion unit” according to the present invention. The conversion table storage unit 103, the voice conversion unit 104, the voice recognition unit 105, and the language model storage unit 106 correspond to the “command word conversion unit” according to the present invention, and the conversion table storage unit 103 and the language model storage unit 106. Corresponds to the “first storage means” and the “second storage means” according to the present invention, respectively. The voice conversion unit 104 and the voice recognition unit 105 correspond to the “second conversion unit” and the “voice recognition unit” according to the present invention, respectively.

  The command signal conversion unit 107 and the command word storage unit 108 correspond to the “command signal generation unit” according to the present invention, and the command signal conversion unit 107 corresponds to the “third conversion unit” according to the present invention. The command word storage unit 108 corresponds to the “third storage unit” according to the present invention.

  The acceleration sensor 101 is connected to the coordinate conversion unit 102 in the apparatus main body 100 via the cable relay member 2 attached to the head of the speaker Q. The cable relay member 2 includes a head mounting portion 201 having elasticity curved in an arc shape, a connection portion 202 between the acceleration sensor 101 provided at both ends of the head mounting portion 201, and one end connected to the connection portion 202. And a signal line 203 connected to the apparatus main body 100 at the other end drawn out from substantially the center of the head mounting portion 201.

  A flexible lead wire 1011 is drawn out from the acceleration sensor 101, and a connector 1012 for electrically connecting to the connection portion 202 is provided at the tip of the lead wire 1011. By connecting the connector 1012 of the acceleration sensor 101 to the connection part 202 of the cable relay member 2, the acceleration sensor 101 is electrically connected to the coordinate conversion part 102 in the apparatus main body 100 by the lead wire 1011 and the signal line 203. . A detection signal of the acceleration sensor 101 is input to the coordinate conversion unit 102 via the lead wire 1011 and the signal line 203.

  As the acceleration sensor 101, as shown in FIG. 2, a small and lightweight semiconductor acceleration sensor having a disk shape with a diameter of several mm is used. The semiconductor acceleration sensor preferably uses, for example, a MEMS (Micro Electro Mechanical Systems) technology.

  The semiconductor acceleration sensor 101 includes a sensor chip that detects acceleration and a signal processing circuit that performs predetermined signal processing on a detection signal output from the sensor chip and outputs the detected signal to the outside. There are various types of acceleration detection methods such as a capacitance type, a piezoresistive type, a gas temperature distribution type, an electrodynamic type, and a magnetic type, and any type of sensor chip may be used.

  The acceleration sensor generally includes a three-axis acceleration sensor that detects acceleration in each axial direction (three directions) of an XYZ orthogonal coordinate system (see FIG. 2), a two-axis acceleration sensor that detects acceleration in two directions (XY direction), There is a one-axis acceleration sensor that detects acceleration in one direction (X direction). As the acceleration sensor 101, a 2-axis acceleration sensor or a 3-axis acceleration sensor is preferably used.

  The acceleration sensor 101 is preferably attached in a pair near both cheeks or both sides of the mouth of the speaker Q, but two or more pairs may be attached, and one in the vicinity of one cheek or one side of the mouth of the speaker Q. It may be attached individually.

  The acceleration sensor 101 is bonded to the cheek of the speaker Q or the peripheral portion of the mouth by, for example, a sheet-like adhesive member 3. For the adhesive member 3, for example, a medical tape or a skin adhesive tape may be used in consideration of adhesion to the skin, adhesive strength, and the like. In the example of FIG. 1, the adhesive member 3 is directly bonded to the acceleration sensor 101, but in order to absorb the vibration of the skin when the speaker Q moves his / her mouth between the acceleration sensor 101 and the adhesive member 3. A buffer member may be interposed.

  In the example of FIG. 1, the acceleration sensor 101R and the acceleration sensor 101L (R is a sensor on the right side of the face of the speaker Q, and L is the position of the face of the speaker Q. The sensor on the left is shown.) Is attached. The acceleration sensors 101R and 101L are triaxial acceleration sensors. By providing a pair of three-axis acceleration sensors 101R, 101L on both sides of the speaker Q's mouth, the mouth when the speaker Q utters speech using the detection signals Sx, Sy, Sz of both acceleration sensors 101R, 101L. Can detect subtle movements.

  FIG. 3 is a diagram illustrating the acceleration detection direction of the acceleration sensor 101 when the acceleration sensor 101 is attached to the face of the speaker Q.

  An XYZ coordinate system that is orthogonal to the acceleration sensor 101 is virtually set in the three-axis acceleration sensor 101, and each axis of the X, Y, and Z axes of the XYZ coordinate system is set from the acceleration sensor 101. Detection signals Sx, Sy, and Sz that detect the acceleration in the direction are output.

  The acceleration sensor 101 matches the X-axis direction and Y-axis direction of the XYZ coordinate system virtually set in the acceleration sensor 101 with the horizontal direction (left-right direction) and vertical direction (vertical direction) of the face of the speaker Q, respectively. It is attached to the cheek of the face of the speaker Q or the peripheral part of the mouth.

  When the acceleration sensor 101 is attached to the cheek of the face of the speaker Q or the periphery of the mouth, the XYZ coordinate system virtually set in the acceleration sensor 101 indicates the acceleration of the face of the speaker Q as shown in FIG. It becomes an orthogonal coordinate system to measure.

  That is, the attachment position of the acceleration sensor 101 on the face of the speaker Q is the origin O of the XYZ coordinate system, and the tangent in the horizontal direction at the origin O of the surface (usually a curved surface) of the speaker Q is the X-axis direction. The tangent of the direction is the Y-axis direction. Further, the normal direction at the origin O of the surface of the face of the speaker Q is the Z-axis direction.

  If the facial expression of the speaker Q whose cheek and muzzle muscles are not moved is a reference facial expression, the position of the origin O of the XYZ coordinate system in the reference facial expression is a detection position for detecting the movement of the speaker's cheek and mouth corner. Pr. When the speaker Q moves the cheek muscles or the muzzle muscles to change the facial expression, the acceleration sensor 101 is displaced according to the change, and the acceleration sensor 101 attached to the speaker Q's face moves the speaker Q. A detection signal Sx that detects the lateral acceleration of the face, a detection signal Sy that detects the vertical acceleration of the face of the speaker Q, and a detection signal Sz that detects the acceleration in the normal direction of the face of the speaker Q. Is output.

  When the acceleration sensor 101 is displaced, as shown in FIG. 4, the detection position Pr for detecting the movement of the cheek or mouth corner of the speaker Q is a position when the reference expression is used (hereinafter, this position is referred to as “reference position”). Displace. The displacement of the detection position Pr from the reference position is represented by a displacement vector V. The components Xd, Yd, and Zd of the displacement vector V in the X-axis, Y-axis, and Z-axis directions are detected signals Sx, Sy, Each can be calculated using Sz.

  The coordinate conversion unit 102 converts the displacement vector V into coordinates Xd, Yd, and Zd in the X-axis direction, the Y-axis direction, and the Z-axis direction using the detection signals Sx, Sy, and Sz output from the acceleration sensor 101. The coordinate conversion unit 102 takes in the detection signals Sx, Sy, Sz from the acceleration sensor 101 at a predetermined period, performs integration processing twice for each detection signal, and detects the distance in each axis direction, that is, the detection position Pr from the reference position. Coordinates Xd, Yd, Zd of the displaced position Pd (hereinafter referred to as “displaced position Pd”) are calculated.

  The conversion table storage unit 103 converts the coordinates Xd, Yd, Zd of the displacement position Pd calculated by the coordinate conversion unit 102 into phoneme data of “A”, “I”, “U”,. The conversion table is stored. The phoneme data is, for example, code data of characters (hirakana or katakana) corresponding to phonemes in a well-known character code table.

  When speaker Q utters a phoneme such as “A”, “I”, “U”, etc., the mouth shape of speaker Q varies depending on the phoneme generated. For example, when vowels of “A”, “I”, and “U” are uttered, the mouth shape is usually as shown in FIGS.

  That is, the shape of the mouth when uttering “a” is a shape with the upper lip and lower lip greatly opened up and down (FIG. 5A), and the shape of the mouth when uttering “i” is the upper lip and With the lower lip slightly opened up and down, the corner of the mouth is pulled sideways (Fig. 5 (b)), and the shape of the mouth when saying "U" is a shape that narrows the mouth so that the lips protrude. (FIG. 5C).

  Since the shape of the mouth of the speaker Q differs depending on the phoneme, when the speaker Q speaks, the position of the acceleration sensor 101 attached to the cheek of the speaker Q or the periphery of the mouth (detection position Pr) when each phoneme is uttered ) Movement (displacement) is different. For example, when the acceleration sensor 101 is mounted directly beside the mouth angle of the speaker Q, for example, the displacement vector V of the detection position Pr when “a” is uttered has a small amount of displacement in the X direction and the Y direction. The amount of displacement in the Z direction tends to increase with respect to the amount of displacement in the X and Y directions. On the other hand, the displacement vector V at the detection position Pr when “I” is uttered tends to have a small amount of displacement in the Y direction and a large amount of displacement in the Y direction relative to the amount of displacement in the X direction.

  Thus, since the displacement vector V of the detection position Pr when the speaker Q utters a phoneme changes characteristically according to each phoneme, the conversion table storage unit 103 stores the displacement vector of the detection position Pr. Data indicating a correspondence relationship between V (coordinates of the displacement position Pd) and phoneme data is stored as a conversion table. This conversion table corresponds to the “first conversion table” according to the present invention.

In the displacement vector V (X, Y, Z) corresponding to the phoneme in the conversion table, for example, a phoneme such as “A”, “I”, “U” is given to the speaker Q who uses the voice command input device 1 in advance. The measured value of the displacement vector V (X, Y, Z) acquired by speaking is used. A large number of speakers Q utter phonemes to measure displacement vectors V (X, Y, Z), and a representative displacement vector V (X, Y, Z) calculated based on the measured values is converted into a conversion table. May be used for the displacement vector V (X, Y, Z) corresponding to the phonemes. Representative displacement vector V (X, Y, Z), for example, measured values for the n displacement vector _{(X m, Y m, Z} m) the mean value of _{(ΣX m / n, ΣY m} / n, ΣZ _m / n).

  FIG. 6 is a diagram illustrating an example of a conversion table stored in the conversion table storage unit 103.

  The conversion table includes the phoneme data “a”, “i”,... “N” and the displacement vector Vi (Xi, Yi, Zi) (i is a code for convenience for distinguishing phonemes. .., I = 1, 2,... In the table of FIG. 6, the “Hiragana” character corresponding to the phoneme is described in the phoneme column, but the code data in the character code table of “Hiragana” or “Katakana” is described in the conversion table. .

  The conversion table shown in FIG. 6 includes data for converting the coordinates Xd, Yd, and Zd of the displacement position Pd into phoneme character codes for 50 phonemes, but for phonemes that are not subject to speech recognition. The conversion data may not be present. For example, when the voice command input device 1 is used in a welding system, command words (hereinafter referred to as “command words”) registered in the voice command of the welding system are limited, and phonemes used for the command words are limited. Therefore, there is no need for conversion data for phonemes that are not used as speech command words.

  For example, the terminology used for welding system voice commands is simple words such as [Kaishi], “Tashishi”, “Denryu”, “Denatsu”, “Give”, “Sageru”, etc. , “Wa”, “Wa” and other phonemes are not used, the conversion table does not have to include conversion data for the phonemes of “Wa” and “Wa”.

  The voice conversion unit 104 refers to the conversion table stored in the conversion table storage unit 103 for the coordinates Xd, Yd, and Zd of the displacement position Pd output from the coordinate conversion unit 102, and the voice data ( Phoneme string to character code string).

  For example, if the coordinates Xd, Yd, Zd of the displacement position Pd output from the coordinate conversion unit 102 are within a predetermined error range (± ΔX, ± ΔY, ± ΔZ) centered on (X1, Y1, Z1). The voice conversion unit 104 converts the phoneme (character code) to “A”. Similarly, if the coordinates Xd, Yd, Zd of the displacement position Pd are (X6, Y6, Z6), the speech conversion unit 104 converts the phoneme (character code) to “ka”. Therefore, when the coordinate data of (X1, Y1, Z1) and (X6, Y6, Z6) is output in order from the coordinate conversion unit 102, the voice conversion unit 104 will generate phonemes (characters) of “a” and “ka”. Code) in order and output.

  The speech recognition unit 105 performs processing for recognizing the character string output from the speech conversion unit 104 using the language model stored in the language model storage unit 106 (recognition processing of speech uttered by the speaker Q). The language model storage unit 106 stores a well-known language model in the speech recognition process. A language model is a formulation of part of speech, syntactic structure, relationship between words, etc. in a Japanese language.

  For example, when the character code string input from the voice conversion unit 104 is “yo”, “u”, “se”, “tsu”, “ka”, “i”, “shi”, the voice recognition unit 105 The voice recognition unit 105 performs a process of recognizing “Yosetsu” + “Kaishi” = “Start welding”.

  The command signal conversion unit 107 performs processing for converting the result of speech recognition output from the speech recognition unit 105 into a command signal using a command word stored in the command word storage unit 108. The command word storage unit 108 stores a conversion table indicating a correspondence relationship between command words registered as voice commands and command signals. This conversion table corresponds to the “second conversion table” according to the present invention.

  The command signal is, for example, an electrical signal that is input to the control unit in order to cause the control unit to control the operation of the actuator. For example, in the case where an operation button “start operation” is provided in the operation unit and a predetermined command signal Sc corresponding to “start operation” is input to the control unit when the operation button is operated, The command signal “command signal Sc” is associated with the command word “start”.

  Therefore, the command signal conversion unit 107 converts the command word stored in the command word storage unit 108 into a command signal when the voice recognition result output from the voice recognition unit 105 is a command word “operation start”. Is used to generate and output a command signal Sc corresponding to the command word.

  As described above, according to the voice command input device 1 according to the present embodiment, the acceleration sensor 101 is attached to the cheek of the speaker (worker) Q or the periphery of the mouth, and the speaker Q utters a voice command. Since the movement of the mouth of the speaker Q is detected using the detection signals Sx, Sy, Sz of the acceleration sensor 101 and the voice command uttered by the speaker Q is recognized from the movement, for example, The voice command uttered by the speaker Q is collected by the microphone, and is hardly affected by noise or noise generated around the speaker Q, compared to the case where the voice is recognized using the collected voice signal. It is possible to recognize a voice command with high accuracy and output a command signal corresponding to the voice command.

  Next, as an application example of the voice command input device 1 according to the present embodiment, a welding system capable of inputting a voice operation command will be described.

  FIG. 7 is a block diagram illustrating a configuration of a welding system including the voice command input device 1.

  A welding system 4 shown in FIG. 7 is a consumable electrode type semi-automatic arc welding system capable of inputting a voice operation command. The welding system 4 includes a welding power source device 5, a wire feeding device 6, a welding torch 7, a gas supply device 8, and a wire reel 9, and the above-described voice for inputting an operation command by voice to the wire feeding device 6. A command input device 1 is provided.

  The welding power source device 5 supplies power (welding power) supplied between the welding wire W and the base material B and power (wire feeding power) supplied to the wire feeding mechanism 601 of the wire feeding device 6. The welding power is generated and supplied to the welding torch 7 via the wire feeding device 6, and the wire feeding power is supplied to the wire feeding mechanism 601 of the wire feeding device 6.

  The welding power supply device 5 includes a power supply unit 501, a control unit 502, an operation unit 503, and a display unit 504 as functional blocks. The power supply unit 501, the operation unit 503, and the display unit 504 are connected to the control unit 502.

  The power supply unit 501 generates drive power for the welding power supply device 5 and also generates drive power for the wire feeding device 6 and welding power. The driving power of the wire feeding device 6 is mainly driving power for driving the control unit 603 and the display unit 605 and driving power for feeding the welding wire W to the welding torch 7 during welding work. The electric power for welding is mainly electric power for generating the arc A between the welding wire W and the base material B and electric power for welding work after the arc is generated.

  The power generation operation of the power supply unit 501 is controlled by the control unit 502. The wire feeding device 6 is connected to the welding power source device 5 by a power cable (not shown) and a control cable (not shown). The driving power of the wire feeding device 6 generated by the power supply unit 501 of the welding power source 5 and the power for welding to the welding torch 7 are supplied to the wire feeding device 6 by a power cable. It is supplied between the welding wire W and the base material B via the feeding device 6. Data communication between the control unit 502 of the welding power source device 5 and the control unit 603 of the wire feeding device 6 is performed via a control cable.

  The power supply unit 501 includes a welding power source 5011 that generates welding power and a wire feeding power source 5012 that generates welding wire feeding power. A welding power source 5011 generates predetermined welding DC power or welding AC power based on a control command of the control unit 502 from a three-phase AC commercial power source P. The welding power changes based on an operation command of a worker during welding or a preset program.

  For example, when the wire feeding mechanism 601 of the wire feeding device 6 feeds the welding wire W with a DC motor, the wire feeding power source 5012 is configured by a known DC power source that generates DC power from the commercial power source P. ing. The wire feeding power source 5012 generates DC power from the commercial power source P, and outputs it to the wire feeding device 6 via the power cable at a driving voltage defined in the wire feeding mechanism 601.

  The power supply unit 501 performs control to supply the shield gas from the gas supply device 8 to the welding torch 7 in synchronization with the power generation operation. The power supply unit 501 is connected to the gas electromagnetic valve 602 in the wire feeder 6 by a control cable. The power supply unit 501 controls the gas electromagnetic valve 602 to the “open” state in synchronization with the start of power generation, and controls the gas electromagnetic valve 602 to the “closed” state in synchronization with the stop of power generation.

  The control unit 502 comprehensively controls processes such as information input by the operation unit 503, information output by the display unit 504, and power supply to the wire feeding device 6 by the power supply unit 501.

  The control unit 502 is realized by, for example, a microcomputer having a ROM (Read Only Memory), a RAM (Random Access Memory), and an MPU (Micro-Processing Unit). Note that the control unit 502 may be realized by a programmable logic device (PLD) such as a field-programmable gate array (FPGA).

  The operation unit 503 is a block that allows an operator to input various information related to arc welding to the control unit 502. The operation unit 503 is provided on, for example, an operation panel disposed on the casing surface of the welding power source device 5. The operation member operated by the worker includes, for example, one or more operation buttons or a power switch arranged on the operation panel for each predetermined item. Note that the operation unit 503 may be configured by other types of input devices such as a touch panel and a keyboard.

  When the operator operates the operation button of the operation unit 503, a predetermined command signal corresponding to the operation button is input to the control unit 502.

  Information on arc welding input by the operator from the operation unit 503 includes, for example, information on a start method, an AC / DC welding method, an initial current, a welding voltage, a welding current, an AC waveform (a pulse waveform or a sine wave). Waveform), information on welding conditions such as frequency in AC welding, and the like. Information relating to arc welding input from the operation unit 503 to the control unit 502 is temporarily stored in a memory (RAM) and transferred to the display unit 504.

  The display unit 504 displays various information related to arc welding so that an operator can visually recognize the information. The display unit 504 is configured by a display device such as an LED or 7 segments, for example. The display unit 504 may be configured by a display device such as a liquid crystal display. The display unit 504 displays information transferred from the control unit 502 in a predetermined display mode.

  The wire feeding device 6 includes a wire feeding mechanism 601, a gas electromagnetic valve 602, a control unit 603, an operation unit 604, a display unit 605, and a voltage / current detection unit 606 as functional blocks. The wire feeder 6 is provided with the voice command input device 1 described above. The operation unit 604, the display unit 605, the voltage / current detection unit 606, and the voice command input device 1 are connected to the control unit 603.

  The wire feeding mechanism 601 is a mechanism for feeding the welding wire W from the wire reel 9 around which the welding wire W is wound during the welding operation and continuously supplying the welding wire W to the welding torch 7. The wire feeding mechanism 601 includes a guide mechanism that guides the welding wire W from the wire reel 9 to the welding torch 7 and a wire feeding motor that is a drive source for feeding the welding wire W. The wire feed motor is driven by the wire feed power supplied from the welding power source device 5.

  The gas electromagnetic valve 602 is a valve for controlling the supply of shield gas from the gas supply device 8 to the welding torch 7. The shield gas is an inert gas such as argon (Ar) or helium (He) for shielding the welding surface from air. The shield gas is sealed in the gas supply device 8 and can be supplied to the welding torch 7 at a predetermined pressure by opening the plug of the gas supply device 8 when the operator starts the welding operation.

  The gas solenoid valve 602 opens the flow path during the period from the start of welding to the end of welding and supplies the shield gas from the gas supply device 8 to the welding torch 7. The opening and closing of the gas electromagnetic valve 602 is controlled by the power supply unit 501 of the welding power supply device 5 as described above.

  The control unit 603 comprehensively controls each process such as input of information by the operation unit 604, output of information by the display unit 605, and feeding of the welding wire W by the wire feeding mechanism 601. The control unit 603 is realized by a microcomputer having a ROM, a RAM, and an MPU. The control unit 603 may also be realized by a PLD such as an FPGA.

  The operation unit 604 is provided on an operation panel disposed on the housing surface of the wire feeding device 6. The operation member operated by the worker includes, for example, one or more operation buttons or a power switch arranged on the operation panel for each predetermined item. Note that the operation unit 604 may also be configured by another type of input device such as a touch panel.

  When the operator operates the operation button of the operation unit 604, a predetermined command signal corresponding to the operation button is input to the control unit 603.

  Information regarding arc welding input by the operator through the operation unit 604 is basically the same as information input by the operator through the operation unit 503 to the control unit 502 of the welding power source device 5. Information regarding arc welding input from the operation unit 604 to the control unit 603 is temporarily stored in a memory (RAM) and transferred to the display unit 605.

  The welding torch 7 holds the welding wire W fed by the wire feeding mechanism 601, and the shielding gas supplied from the gas supply device 8 via the gas electromagnetic valve 602 is used as the shielding gas by the welding wire W and the mother wire. It is an apparatus that fulfills the function of outputting so as to cover the arc A generated between the material B and the weld metal. The welding torch 7 is provided with a torch switch 701 for an operator to adjust the welding current at hand.

  The torch switch 701 is connected to the control unit 603 of the wire feeding device 6 by a control cable (not shown). The torch switch 701 is constituted by, for example, a push button type momentary switch. For example, a high level signal is input to the control unit 603 when the torch switch 701 is “ON”, and a low level signal is input when the torch switch 701 is “OFF”.

The operator can input a pulse signal having a pulse width t to the control unit 603 by turning the torch switch 701 ON after the predetermined time t has elapsed. The control unit 603 is provided with a table for converting the waveform of the pulse signal input from the torch switch 701 into a specific operation command.

  For example, the control unit 603 assigns a click operation in which a pulse width t is a pulse signal larger than a predetermined time to an operation command for energizing or interrupting a welding current, and a pulse signal having a pulse width t within a predetermined time. A command to increase the welding current in a predetermined step by assigning the single click operation to a command signal for decreasing the welding current in a predetermined step and generating a double-click operation in which the same pulse signal is generated twice continuously within a predetermined time interval. A table assigned to the signals is provided.

  By operating the torch switch 701 with a predetermined operation pattern, the operator can input commands such as the start of the welding operation, the change of the welding current during welding, and the end of the welding operation to the control unit 603. When a command signal is input from operation unit 604, control unit 603 transmits the command signal to control unit 502 of welding power supply device 5.

  In the present embodiment, the torch switch 701 is connected to the control unit 603 of the wire feeding device 6, but the torch switch 701 is connected to the control unit 502 of the welding power source device 5 to control the operation signal of the torch switch 701. The data may be input directly to the unit 502.

  The voltage / current detection unit 606 detects a voltage (welding voltage) between the welding wire W and the base material B and a current (welding current) flowing between the welding wire W and the base material B. The voltage / current detection unit 606 converts a voltage sensor that detects a voltage and a voltage detection value output from the voltage sensor into a digital signal and outputs the digital signal to the control unit 603. In addition, the voltage / current detection unit 606 converts a current sensor that detects current and a current detection value output from the current sensor into a digital signal and outputs the digital signal to the control unit 603.

  Information on arc welding input by the operator from the operation unit 604 or the torch switch 701 and information on a welding voltage detection value and a welding current detection value during welding work are input from the voltage / current detection unit 606 to the control unit 603. Is done. These pieces of information input to the control unit 603 are transmitted from the wire feeding device 6 to the welding power source device 5 by data communication via the control cable.

  The display unit 605 is configured by a display device such as an LED or a 7 segment. The display unit 605 may be configured by a display device such as a liquid crystal display. The display unit 605 performs a predetermined display based on information transferred from the control unit 603.

  The voice command input device 1 includes a coordinate conversion unit 102, a conversion table storage unit 103, a voice conversion unit 104, a voice recognition unit 105, and a language model storage unit provided in the device main body 100 of the voice command input device 1 shown in FIG. 106, a command signal conversion unit 107 and a command word storage unit 108 are provided in the wire feeding device 6, and the acceleration sensor 101 of the voice command input device 1 is attached to the cheek of the worker Q or the peripheral part of the mouth. . In FIG. 7, a voice command input unit 109 indicates a portion provided in the device main body 100 of the voice command input device 1.

  In the wire feeding device 6, the operator can input an operation command with either the operation unit 604 or the voice command input device 1. For this reason, the operation panel of the wire feeding device 6 is provided with a selection button for selecting a device for inputting an operation command. This selection button is composed of, for example, a momentary switch, and an operation command can be input to the control unit 603 by voice using the voice command input device 1 during a period in which the worker is turning on the momentary switch. .

  Since the acceleration sensor 101 and the coordinate conversion unit 102 in the voice command input device 1 are as described above, the voice conversion unit 104 to the command word storage unit 108 will be described below.

  Since the word used for the voice command in the welding system 4 is a part of the 50 sounds, a conversion table is created in advance and stored in the conversion table storage unit 103 for the phonemes used in the voice command. . For example, if the term registered in the voice command is “Denryu”, “Denatsu”, “Guru”, “Sasaru”, “Yosetsu”, “Kaishi”, “Toshishi”, etc. / De "," n "," ryu "," u "," tsu "," a "," ke / ge "," ru "," ka "," sa "," yo "," se " , “I”, “Shi”, and other phoneme conversion tables are stored in the conversion table storage unit 103.

  The voice conversion unit 104 utters the operator Q with reference to the conversion table stored in the conversion table storage unit 103 for the column of coordinates (Xd, Yd, Zd) of the displacement position Pd input from the coordinate conversion unit 102. It is converted into speech data (data obtained by converting a phoneme string into a character code string).

  For example, when the coordinates (Xd, Yd, Zd) are (X19 ± ΔX, Y19 ± ΔY, Z19 ± ΔZ) (see FIG. 6), the voice conversion unit 104 converts the phoneme (character code) to “te”. To do. However, when the coordinates (Xd, Yd, Zd) are not stored in the conversion table storage unit 103, the voice conversion unit 104 converts the code into a code of a symbol that is not a character such as “?”.

  For example, from the coordinate conversion unit 102 to the voice conversion unit 104, (X19 ± ΔX, Y19 ± ΔY, Z19 ± ΔZ), (X2 ± ΔX, Y2 ± ΔY, Z2 ± ΔZ), (X12 ± ΔX, Y12 ± ΔY, Z12 ± ΔZ) (refer to FIG. 6), when the coordinate data is input in order, there is conversion data for converting these coordinates into phonemes. It is converted into “shi” phonemes in order and output.

  However, in the case where the second coordinate data input to the voice conversion unit 104 is a phoneme that is not stored in the conversion table storage unit 103, for example, a “ni” phoneme that resembles the mouth shape of “i”, The voice conversion unit 104 converts the phonemes into “te”, “?”, And “shi” phonemes in order and outputs them.

  The speech recognition unit 105 performs a process of recognizing speech using the language model stored in the language model storage unit 106 from the character code string input from the speech conversion unit 104. The language model storage unit 106 stores terminology and word data used in command words in the welding system.

  FIG. 8 is a diagram illustrating an example of term and word data used for the command word stored in the language model storage unit 106.

  As shown in FIG. 8, the language model storage unit 106 stores “welding”, “voltage”, “current”, “raising”, “lowering”, “starting”, “stopping”, “high”, “low”. , “Yes”, “No”, etc., terms and words used for instructions in welding work are registered.

  In the language model storage unit 106, welding terms and words having two or more characters are registered in hiragana or katakana, but one-letter terms and words may be registered. In the following description, for convenience of explanation, it is assumed that two or more characters or words for welding are registered in the language model storage unit 106.

  Also, if the term or word used in the command includes a homonym term or word, one term is changed to a synonym or synonym and registered in the language model storage unit 106 so that the character string is different. It shall be. For example, if the terms used in the command include the terms “machining” and “descent”, the term “descent” is changed to the term “descent”. It is registered in the language model storage unit 106. As a result, the language model storage unit 106 registers terms and words used in the command so as not to include homonym terms as much as possible.

  The voice recognition unit 105 collates a character string (character code string) input from the voice conversion unit 104 with a term (hereinafter referred to as “registered term”) stored in the language model storage unit 106. A term whose character string completely matches is recognized as a term uttered by the speaker Q. For example, the character string input from the voice conversion unit 104 is “de”, “n”, “a”, “tsu”, and the term “voltage” is registered in the language model storage unit 106. If so, the voice recognition unit 105 performs a process of recognizing “denatsu” = “voltage”.

  In addition, when the character string input from the voice conversion unit 104 does not completely match the registered term stored in the language model storage unit 106, the voice recognizing unit 105 determines whether the character string has the same number of characters as the character string. The degree of coincidence E is calculated, and voice recognition processing is performed using the degree of coincidence E.

  In the present embodiment, for example, the number of characters of the character string input from the speech conversion unit 104 is n (integer), and the number of characters that match the registered term character among the characters of the character string is r (1 ≦ r ≦ n−1), the character string matching degree E is defined as E = r / n (<1) or E = r × 100 / n (<100) [%]. For example, when the number of characters in the character string input from the voice conversion unit 104 is four and the number of characters that match a registered term consisting of four characters is three, the degree of match E is 3/4 = 0.75. Or 75 [%].

When there are two or more registered terms having the same number of characters, the speech recognition unit 105 calculates the degree of coincidence E for each registered term, and the degree of coincidence is the highest among registered terms having a coincidence E equal to or higher than a preset threshold E _TH A registered term having a high E is selected as a recognition target term.

The threshold E _TH can be arbitrarily set within the range of 0.5 <E _TH <1.0. However, in the present embodiment, for example, for a registered term having an odd number of characters n, the threshold E _TH1 = 2 / 3≈0.67 is set, and a threshold value E _TH2 = 3/4 = 0.75 is set for a registered term having an even number of characters n.

The threshold value E _TH1 = 2/3 for a registered term with an odd number of characters n corresponds to the degree of coincidence E when a majority of characters (2 characters) match in a registered character of 3 characters. In the case of a registered term of 3 characters, if there is a match E of 2/3 or more, there is only a case where 2 characters match, so 2 characters out of 3 character strings input from the speech converter 104 match. The registered term is selected as the recognition target term. In addition, in the registered term of 5 characters, if the matching degree E is 2/3 or more, there is only a case where 4 characters are matched, so 4 characters out of 5 character strings input from the voice conversion unit 104 are matched. The registered term is selected as the recognition target term.

  Therefore, when the character string input from the voice conversion unit 104 is three characters, the voice recognition unit 105 recognizes a registered term that matches two characters in the character string as a term uttered by the speaker Q. In addition, when the character string input from the voice conversion unit 104 is five characters, the voice recognition unit 105 recognizes a registered term that matches four characters in the character string as a term uttered by the speaker Q.

  For 7-character registered terms, registered terms that have a match E of 2/3 or more are registered terms that match 5 or 6 characters. However, the 6-character has the highest match E. Matching registered terms. Therefore, the speech recognition unit 105 recognizes the registered term as a term uttered by the speaker Q if there is a registered term that matches 6 characters in the 7-character string input from the speech conversion unit 104. If there is only a registered term that matches the character, the registered term is recognized as a term uttered by the speaker Q.

  In the case of a registered term of 9 characters or more, the speech recognition unit 105 registers the registered term having the highest matching degree E among the registered terms having a matching degree E of 2/3 or more, similarly to the case of the registered term of 7 characters. Is recognized as a term uttered by speaker Q.

  For example, when a character string “de”, “n”, “a”, “tsu”, “shi” is input to the speech recognition unit 105, the language model storage unit 106 uses 2/3 characters in terms of 5 characters. If only the registered term “denatsuchi” = “voltage value” is registered, the speech recognition unit 105 recognizes the character string “denatsushi” as “voltage value”. . In other words, the voice recognition unit 105 recognizes that “shi” in the character string “denatsushi” is “denatsuchi (voltage value)” by regarding it as an erroneous conversion of the phoneme of “chi”.

  If there are a plurality of recognition target terms, the speech recognition unit 105 recognizes any one term as a term uttered by the speaker Q in consideration of the relationship with terms before and after the term. For example, a character string “de”, “n”, “ryu”, “u”, “e”, “i”, “shi” (“e” is an erroneously converted character) is stored in the voice recognition unit 105. When the voice recognition unit 105 is input, the character strings “de”, “n”, “ryu”, and “u” are recognized as “denryu” = “current”, and “e”, “i”, “ It is assumed that “kaishi” = “start” and “deashi” = “stop” are selected as recognition targets of the character string “shi”.

  “Start” is a command word “current start” that has no meaning for “current”, whereas “stop” is a command word “current stop” that has a meaning for “current”. Therefore, the voice recognition unit 105 recognizes the character strings “e”, “i”, and “shi” as “deshi” = “stop”.

The threshold E _TH2 = 3/4 for a registered term having an even number of characters n corresponds to the degree of coincidence E when a majority of characters (3 characters) match in a registered character of 4 characters.

The two-character registered term is a case where the majority of the characters match, and in this case, the voice recognition is possible. Therefore, the voice recognition unit 105 performs a normal voice recognition process. On the other hand, if any one of the two character strings input from the voice conversion unit 104 matches the registered term character, or if any of the two character strings does not match the registered term, they match. Since the degree E does not satisfy the condition of the threshold value E _TH2 = 3/4 or more, correct speech recognition cannot be performed. Therefore, in these cases, the speech recognition unit 105 performs a process that cannot be recognized.

  In the case of a registered term of 4 characters, if there is a match E of 3/4 or more, there is only a case where 3 characters match, so 3 characters in the 4 character string input from the voice conversion unit 104 match. The registered term is selected as the recognition target term. In addition, in the registered term of 6 characters, if the matching degree E is 3/4 or more, there is only a case where 5 characters are matched, so 5 characters in the 6-character string input from the voice conversion unit 104 are matched. The registered term is selected as the recognition target term.

  Therefore, when the character string input from the voice conversion unit 104 is four characters, the voice recognition unit 105 recognizes a registered term that matches three characters in the character string as a term uttered by the speaker Q. When the character string input from the voice conversion unit 104 is six characters, the voice recognition unit 105 recognizes a registered term that matches five characters in the character string as a term uttered by the speaker Q.

  For registered terms with 8 characters, registered terms with a matching degree E of 3/4 or higher are registered terms that match 6 or 7 characters, but 7 characters have the highest degree of matching E. Matching registered terms. Therefore, the speech recognition unit 105 recognizes the registered term as a term uttered by the speaker Q if there is a registered term that matches 7 characters in the 8-character string input from the speech conversion unit 104. If there is only a registered term that matches the character, the registered term is recognized as a term uttered by the speaker Q.

The voice recognition unit 105 performs the same recognition process as in the case where the character string described above has an odd number of characters even when the character string input from the voice conversion unit 104 has an even number of characters. In addition, when there is no registered term having a matching degree E equal to or higher than the threshold value E _{TH with} respect to the character string input from the voice conversion unit 104, the voice recognition unit 105 performs a process that cannot be recognized. The unrecognizable process is, for example, a process of not outputting a recognition result from the voice recognition unit 105 to the command signal conversion unit 107 or a process of outputting data of a predetermined recognition result that cannot be converted into a command signal to the command signal conversion unit 107.

  For example, when processing that does not output the recognition result from the speech recognition unit 105 to the command signal conversion unit 107 as unrecognizable processing, the command signal conversion unit 107 does not receive recognized command word data. The unit 107 does not perform conversion processing into a command signal. Further, when data of a predetermined recognition result is output from the voice recognition unit 105 to the command signal conversion unit 107 as a process that cannot be recognized, the data of the command signal corresponding to the data of the predetermined recognition result is stored in the second conversion table. In this case as well, the command signal conversion unit 107 does not substantially convert the command signal.

In the above example, the threshold value E _TH1 for a registered term having an odd number of characters and the threshold value E _TH2 for a registered term having an even number of characters are fixed to one regardless of the number of characters, but the threshold value _TH depends on the number of characters of the registered term. May be different.

  Further, in the above example, when there are a plurality of registered terms that are sentence recognition targets, any one registered term is recognized as a term uttered by the speaker Q in consideration of the relationship with terms before and after the registered term. However, it is also possible to recognize a term uttered by the speaker Q by estimating erroneous conversion of characters that do not match each registered term.

  For example, the voice recognition unit 105 selects “Kaishi” = “Start” and “Toshi” = “Stop” as recognition targets for the three character strings “E”, “I”, and “Shi”. In the example, the voice recognition unit 105 performs a process of estimating an erroneous conversion for a character “E” that does not match the character string “EISHI” and the registered terms “KAIISHI” and “TAISEI”. For example, the voice recognition unit 105 extracts the vowels of the characters “e”, “ka”, and “te”, and among the characters “ka” and “te”, the vowel close to the mouth shape of “e” The character “te” having “is selected.

  The speech recognition unit 105 estimates that the character “e” is an erroneous conversion of the character “te”, and among the registered terms selected as recognition targets, the speaker Q selects the registered term “deshi” = “stop”. Recognize a spoken term.

  The command signal conversion unit 107 performs processing for converting the result of speech recognition output from the speech recognition unit 105 into a command signal. The command word storage unit 108 stores a conversion table indicating the correspondence between command words registered as voice commands in the welding system and command signals. The command signal conversion unit 107 converts the speech recognition result output from the speech recognition unit 105 into a command signal using the conversion table stored in the command word storage unit 108 and inputs the result to the control unit 603.

  The voice command is a command corresponding to a part of a plurality of operation commands input to the control unit 603 by the operation button of the operation unit 604 or the torch switch 701. As the voice command, an operation command that is highly necessary to be input to the welding power source device 5 by the operator during the welding work is selected. Moreover, the term for welding is a word which comprises an operation command.

  FIG. 9 is a diagram showing an example of a command word of the welding system registered in the command word storage unit 108 and its contents.

  As shown in FIG. 9, the command word storage unit 108 has welding operations such as “welding start”, “welding stop”, “voltage increase”, “voltage decrease”, “current increase”, “current increase”, etc. The command word is registered. For example, the command word “start welding” or “stop welding” is a command word for instructing to start or stop supplying welding power (arc generation power and welding power) to the welding torch 7 and the base material B. It is.

  Moreover, the command word “voltage increase” or “voltage decrease” increases or decreases the welding voltage by a unit voltage value ΔV (for example, 0.1 [V]) preset with respect to the current welding voltage value. It is a command to make. Further, the command word “current increase” or “current decrease” increases or decreases the welding current by a unit current value ΔI (for example, 1.0 [A]) set in advance with respect to the current welding current value. It is a command to be made.

  For example, when the recognition result of “welding start” is input from the voice recognition unit 105, the command signal conversion unit 107 generates a command signal corresponding to the “welding start” command and inputs the command signal to the control unit 603. Further, for example, when the recognition result of “voltage increase” or “voltage decrease” is input from the voice recognition unit 105, the command signal conversion unit 107 responds to a command of “increase welding voltage” or “decrease welding voltage”. A command signal is generated and input to the control unit 603.

  On the other hand, if the command word corresponding to the recognition result input from the speech recognition unit 105 is not included in the conversion table of the command word storage unit 108, the command signal conversion unit 107 does not have a command signal corresponding to the recognition result. , Do not process change to command signal. In addition, the command signal conversion unit 107 does not perform the process of changing to a command signal even when a recognition result is not input from the voice recognition unit 105 as a process that cannot be recognized. Therefore, in these cases, no command signal is input to the control unit 603.

  When the voice command uttered by the worker Q cannot be recognized, the command signal is not input to the control unit 603. Therefore, the worker Q utters the voice command until the command signal is input to the control unit 603. Will be repeated.

  When the command signal is input from the command signal conversion unit 107, the control unit 603 transmits the command signal to the control unit 502 of the welding power source device 5 through the control cable. When a command signal is input from the control unit 603 of the wire feeder 6, the control unit 502 executes a predetermined process corresponding to the command signal.

  For example, when the control unit 603 receives a command signal “increased welding current”, the control unit 603 outputs a welding current output from the welding power source 5011 of the power source unit 501 to a predetermined current value (for example, 1 Control to increase only by [A]) is performed.

  Next, voice command input control by the voice command input device 1 will be described with reference to the flowchart of FIG.

  In the following description, the operator operates the selection button of the operation unit 604 during the welding operation to place the wire feeding device 6 in the voice command input permission state, and uses the voice command input device 1 to send the voice command to the control unit 603. The case of inputting to will be described. In addition, a voice command “increase welding current” for increasing the welding current by a predetermined current value will be described as an example of the voice command input by the operator during the welding operation.

  The control unit 603 monitors whether or not a voice command input permission has been input from the operation unit 604 (loop processing of S100). If the voice command input permission is not input (S100: N), the control unit 603 maintains the normal input state in which the operation command is input by the operation unit 604 or the torch switch 701, and the voice command input permission is input. If so (S100: Y), the normal input state is switched to the voice command input permission state, and the process proceeds to step S101 to perform voice command input processing.

  In step S101, the coordinate conversion unit 102 receives the acceleration detection signals Sx, Sy, Sz in the X-axis, Y-axis, and Z-axis directions output from the acceleration sensor 101, and the detection signals Sx, Sy. , Sz are converted into coordinates (Xd, Yd, Zd) of the displacement position Pd displaced from the detection position Pr.

  Subsequently, the phonetic data (character code data) with reference to the conversion table stored in the conversion table storage unit 103 is the coordinates (Xd, Yd, Zd) output from the coordinate conversion unit 102 by the voice conversion unit 104. (S102).

  Subsequently, the speech recognition unit 105 recognizes speech using the language model stored in the language model storage unit 106 for the character string data (character code string data) output from the speech conversion unit 104. Processing is performed (S103), and it is determined whether or not the command word has been recognized by the voice recognition processing (S104).

  The speech recognition unit 105 collates the character string input from the speech conversion unit 104 with the registered term stored in the language model storage unit 106, and if there is a registered term whose character string completely matches, the registration is performed. The term or combination of registered terms is recognized as a command word uttered by the speaker Q, and the recognition result (command word data) is output to the command signal conversion unit 107 (S104: Y), and the process proceeds to step S108.

  For example, if the character string output from the voice conversion unit 104 is “de”, “n”, “ryu”, “u”, “a”, “ge”, “ru”, the language model storage unit 106 If the registered terms "denryu" and "raise" are registered, the speech recognition unit 105 recognizes the character string as "denryu" + "raise" = "current increase" command word, The “current increase” data is output to the command signal converter 107 (S104: Y), and the process proceeds to step S108.

  On the other hand, when the registered term that completely matches the character string input from the speech conversion unit 104 is not stored in the language model storage unit 106 (S104: N), the speech recognition unit 105 stores it in the language model storage unit 106. Among the registered terms, the degree of coincidence E is calculated for the registered terms having the same number of characters as the character string input from the speech conversion unit 104 (S105).

  Then, the voice recognition unit 105 performs voice recognition processing using the degree of coincidence E (S106), and determines whether or not the command word can be recognized by the voice recognition processing (S107).

That is, for each registered term for which the matching degree E is calculated, the voice recognition unit 105 _{selects the} registered term having the highest matching degree E among the registered terms having a matching degree E equal to or higher than a preset threshold E _TH. Is recognized and output to the command signal converter 107 (S107: Y), and the process proceeds to step S108.

In addition, when there are two or more registered terms having the highest matching degree E among the registered terms having the matching degree E equal to or higher than the threshold E _TH , the voice recognition unit 105, for example, recognizes registered registrations before and after each registered term. One registered term is determined in consideration of the relationship with the term, and a combination of the registered term and the preceding and following registered terms is recognized as a command word uttered by the speaker Q.

  For example, the speech recognition unit 105 excludes the selected registered term from the recognition target when the selected registered term becomes a command word that does not make the meaning of the command as a whole in relation to the previous and subsequent recognized registered terms. When the command word makes sense, the combination of the selected registered term and the previous and subsequent registered terms is recognized as the command word uttered by the speaker Q.

  For example, the voice recognition unit 105 recognizes “denryu” = “current” for the character string “denrye” inputted from the voice conversion unit 104, and “ Suppose you select registered terms for “stop” and “start”. The voice recognizing unit 105 recognizes “deniyuusei” as “current stop” because “current start” does not make sense as a command word, but “current stop” makes sense as a command word.

  Then, the voice recognition unit 105 outputs the recognized command word data to the command signal conversion unit 107 (S107: Y), and proceeds to step S108.

On the other hand, if there is no registered term having a matching degree E equal to or higher than the threshold E _{TH with} respect to the character string input from the voice conversion unit 104, the voice recognition unit 105 sends an unrecognized processing result to the command signal conversion unit 107. Output (S107: N) and return to step S100.

  In step S108, the command signal conversion unit 107 performs a process of converting the voice recognition result output from the voice recognition unit 105 into a command signal using the command word stored in the command word storage unit 108. Do.

  When the voice recognition result (command word data) is included in the conversion table of the command word storage unit 108, the command signal conversion unit 107 generates a command signal corresponding to the voice recognition result and outputs the command signal to the control unit 603. After (S108), the process returns to step S100. For example, when the voice recognition result output from the voice recognition unit 105 is “current increase”, the command signal conversion unit 107 uses the command word “current increase” (the welding current is increased by the unit current value ΔI [A]). Is generated and input to the control unit 603, and then the process returns to step S100.

  The command signal conversion unit 107 generates a command signal when the voice recognition result is not included in the conversion table of the command word storage unit 108 or when the voice recognition result is not input from the voice recognition unit 105 due to the recognition failure. It returns to step S100, without doing.

  As described above, according to the welding system 4 according to the present embodiment, since the voice command input device 1 is used, even when an operator inputs a command by voice during the welding work, it occurs at the work site. The command signal corresponding to the voice command can be input stably with high accuracy without being affected by very high noise.

  Further, since the acceleration sensor 101 is directly attached to the cheek or the peripheral part of the mouth of the worker Q and the voice uttered by the worker Q is detected using the detection signals Sx, Sy, Sz of the acceleration sensor 101, the worker Q As compared with the configuration in which the microphone is provided close to the mouth, the sensor does not get in the way, and the adverse effect on the welding work of the worker Q can be reduced.

  In addition, since a pair of three-axis acceleration sensors 101 are provided on both sides of the mouth of the worker Q so as to detect the movement of the mouth, the mouth movement that slightly changes according to the voice uttered by the worker Q. Can be detected with high accuracy.

  Further, since the acceleration sensor 101 is directly attached to the face of the worker Q and the acceleration sensor 101 and the apparatus main body 100 are electrically connected via the cable relay member 2, the acceleration sensor 101 is welded by the worker Q. The workability will not be deteriorated.

  Further, even when the character string of the voice command input to the voice recognition unit 105 does not completely match the registered term stored in the language model storage unit 106, for example, the matching degree E between the character string and the registered term character is determined. The command word having a high probability of matching with the character string of the voice command is calculated using the matching degree E, and is recognized as the voice command uttered by the worker Q. The voice command can be suitably recognized using the detected value of the movement.

  Further, even when a character obtained by converting the phoneme of the voice command by the voice conversion unit 104 includes erroneous conversion, the voice recognition unit 105 calculates the degree of coincidence E with the registered term, and uses the degree of match E to generate the voice command. Therefore, the command signal corresponding to the voice command uttered by the operator Q can be input to the control unit 603 of the wire feeder 6 with high accuracy.

  In the above embodiment, the operation unit 604 is provided with a selection switch so that an operator can input a voice operation command by operating the selection switch. However, instead of the selection switch, a mode changeover switch is provided in the operation unit 604. It may be provided.

  The mode switch is a switch for switching the operation command input mode between a voice input mode (voice input mode) and an operation button or torch switch 701 input mode (normal input mode).

  When the operator sets the voice input mode with the mode change switch, the control unit 603 performs reception control of the operation command signal by the operation button of the operation unit 604 or the torch switch 701, and the operation command signal from the voice command input device 1. The admission control is performed. Conversely, when the operator sets the normal input mode with the mode change switch, the control unit 603 accepts an operation command signal from the operation unit 604 operation button or the torch switch 701, but receives the operation command signal from the voice command input device 1. Control not accepted.

  The welding system 4 shown in FIG. 7 does not include a remote operation device, but may include a remote operation device.

  FIG. 11 is a diagram illustrating an example of a welding system field configuration including a remote control device.

  A welding system 4A shown in FIG. 11 includes a remote operation device 10 that can be connected to a wire feeding device 6A by wireless communication. The remote operation device 10 includes a voice command input device 1.

  In the welding system 4 shown in FIG. 7, the voice command input device 1 is provided in the wire feeding device 6 arranged near the worker Q. However, in the welding system 4A shown in FIG. Since the operation device 10 is provided, the arrangement of the voice command input device 1 to the welding system 4A can be made easier and more compact.

  In particular, when the voice command input device 1 is provided in the wire feeding device 6, wiring from the acceleration sensor 101 to the device main body 100 of the voice command input device 1 in consideration of the movement of the worker Q from the wire feeding device 6. However, when the voice command input device 1 is provided in the remote control device 10, the distance between the worker Q and the remote control device 10 is short and the change in the distance is small. The wiring to the apparatus main body 100 of the command input device 1 can be made more compact.

  The difference in configuration of the welding system 4A from the welding system 4 shown in FIG. 7 will be briefly described as follows.

  In the welding system 4A, the wire feeder 6A includes a wireless communication unit 607. The wireless communication unit 607 performs wireless communication with the remote control device 10 by, for example, a wireless LAN. The wireless communication unit 607 wirelessly connects to the remote control device 10 via a wireless LAN access point (not shown), and transmits / receives predetermined information by wireless communication. The wireless communication unit 607 modulates the carrier signal with information to be transmitted using a predetermined modulation method to generate a wireless communication signal, and radiates the wireless communication signal from the antenna 607A.

  For example, when information on the welding voltage and welding current is transferred from the control unit 603, the wireless communication unit 607 generates a wireless communication signal using the information and transmits the wireless communication signal to the remote control device 10. In addition, when the information on the welding voltage detection value and the welding current detection value is transferred from the control unit 603, the wireless communication unit 607 generates a wireless communication signal using the information, and the wireless communication signal is transmitted to the remote control device 10. Send to.

  In addition, when receiving a wireless communication signal from the remote control device 10, the wireless communication unit 607 demodulates the transmitted information and transfers it to the control unit 603. Further, the wireless communication unit 607 detects the reception level and inputs the detected value to the control unit 603. In order to display the communication state of the wireless communication, the control unit 603 generates information related to the reception level based on the detected value, and transfers the information to the display unit 605.

  The remote control device 10 is called a remote controller. The remote operation device 10 includes, as functional blocks, a battery 1001, a control unit 1002, a wireless communication unit 1003, an operation unit 1004, a display unit 1005, and the voice command input unit 109 of the voice command input device 1 described above.

  The battery 1001 is a drive power source for the remote operation device 10. The battery 1001, the wireless communication unit 1003, the operation unit 1004, the display unit 1005, and the voice command input unit 109 are connected to the control unit 1002.

  The control unit 1002 performs various processes such as information input by the operation unit 1004, information output by the display unit 1005, voice command input by the voice command input unit 109, and data communication with the wire feeder 6A by the wireless communication unit 1003. Overall control. The control unit 1002 is realized by a microcomputer having a ROM, a RAM, and an MPU. The control unit 1002 may also be realized by a PLD such as an FPGA.

  The operation unit 1004 is a block for performing an operation in which an operator inputs various types of information related to arc welding to the control unit 1002. The operation unit 1004 is provided on an operation panel disposed on the housing surface of the remote operation device 10, and performs the same function as the operation unit 604 of the wire feeding device 6A.

  When the operator operates an operation button such as starting the operation unit 1004, setting / changing the welding voltage, setting / changing the welding current, etc., a predetermined operation command signal corresponding to the operation button is input to the control unit 1002.

  In the remote operation device 10, the operator can input an operation command with either the operation unit 1004 or the voice command input device 1. For this reason, the operation panel of the remote operation device 10 is provided with a selection button for selecting a device for inputting an operation command. This selection button is composed of, for example, a momentary switch, and an operation command can be input to the control unit 1002 by voice using the voice command input device 8 during a period in which the worker is turning on the momentary switch. .

  The display unit 1005 displays various information related to arc welding, information related to the remaining capacity of the battery 1001, the state of wireless communication with the wire feeding device 6A, and the like. The display unit 1005 is configured by a display device such as a liquid crystal display, for example. The display unit 1005 may be configured by a display device such as an LED or 7 segments.

  The display unit 1005 performs a predetermined display based on information transferred from the control unit 1002. For example, the display unit 1005 displays information on the remaining capacity of the battery 1001 with pictographs, symbols, character messages, and the like based on information on the voltage of the battery 1001 input from the control unit 1002. Further, the display unit 1005 displays information on the communication state of wireless communication as pictographs, symbols, text messages, and the like based on information on the reception level input from the control unit 1002.

  The display unit 1005 also receives input information transferred from the control unit 1002 (information input from the operation unit 1004 or the voice command input unit 109 or information input from the wire feeder 6A via the wireless communication unit 1003). Is displayed.

  The wireless communication unit 1003 performs wireless communication with the wire feeding device 6A. Similarly to the wireless communication unit 34, the wireless communication unit 1003 performs wireless communication with the wire feeding device 6A through a wireless LAN.

  The wireless communication unit 1003 wirelessly connects to the wire feeding device 6A via a wireless LAN access point (not shown), and transmits / receives predetermined information by wireless communication. The wireless communication unit 1003 generates a wireless communication signal by modulating a carrier signal with information to be transmitted using the same modulation method as the wireless communication unit 607, and radiates the wireless communication signal from the antenna 1003A.

  For example, when information on the welding voltage and welding current is transferred from the control unit 1002, the wireless communication unit 1003 generates a wireless communication signal using the information and transmits the wireless communication signal to the wire feeding device 6A. .

  Further, when receiving the wireless communication signal from the wire feeding device 6 </ b> A, the wireless communication unit 1003 demodulates the transmitted information and transfers it to the control unit 1002. Radio communication unit 1003 detects the reception level and inputs the detected value to control unit 1002. In order to display the communication state of the wireless communication, the control unit 1002 generates information on the reception level based on the detected value, and transfers the information to the display unit 1005.

  In the welding system 4A, when the voice command uttered by the worker Q is converted into a command signal by the voice command input device 1 and input to the control unit 1002 of the remote operation device 10, the control unit 1002 corresponds to the command signal. Information on the operation command is transmitted to the control unit 603 of the wire feeding device 6A by wireless communication, and the information on the operation command is further transferred from the control unit 603 to the control unit 502 of the welding power source device 5.

  As a result, the welding power supply device 5 performs a predetermined operation corresponding to the information of the operation command input from the wire feeding device 6A. For example, if the content of the operation command is “increase in welding current”, the welding power source device 5 performs control to increase the welding current supplied to the welding torch 7 and the base material B by the welding power source 5011 by ΔI. Is called.

  In the configuration of FIG. 11, the remote operation device 10 performs wireless communication with the wire feeding device 6 </ b> A, but the remote operation device 10 performs wireless communication with the welding power source device 5. The wireless power communication may be performed with respect to both the welding power supply device 5 and the wire feeding device 6A.

  As described above, according to the welding systems 4 and 4A according to the present embodiment, the movement of the cheek or the mouth when the worker Q issues a voice command is detected by the acceleration sensor 101, and based on the detection result. Since the content of the voice command is recognized and the command signal corresponding to the voice command is input to the control unit 502 of the welding power source device 5, the noise is affected even in a noisy environment at the welding work site. The contents of the voice command can be recognized without any problem.

  In particular, in the welding systems 4 and 4A, the operation command such as changing the welding condition by the operation button has relatively simple contents, and the voice command word is composed of simple and few phonemes. Malfunctions and misrecognition can also be reduced by voice command recognition processing using the detected value of the movement of the cheek or mouth of Q.

  Even if a part of the phoneme of the voice command obtained from the detected value of the movement of the cheek or mouth of the worker Q by the acceleration sensor 101 is erroneously detected, for example, the matching degree E between the character string and the character of the registered term is determined. Since the coincidence E is calculated and a registered term or a combination of registered terms having a high probability of matching the character string of the voice command is recognized as the voice command uttered by the worker Q, the voice command is high. It can be converted into a command signal corresponding to the voice command and input with accuracy.

  Therefore, the command signal of the investigation command regarding welding can be input to the welding power source device 5 stably with high accuracy by the voice command. Thereby, the operator Q can reduce the manual input operation for changing the welding conditions during the welding operation, and the welding operation is facilitated.

DESCRIPTION OF SYMBOLS 1 Voice command input device 100 Apparatus main body 101,101R, 101L Acceleration sensor 1011 Lead wire 1012 Connector 102 Coordinate conversion part (1st conversion means)
103 Voice conversion unit (second conversion means)
104 Conversion table storage unit (first storage unit)
105 Voice recognition unit (voice recognition means)
106 Language model storage unit (second storage means)
107 Command signal converter (third converter)
108 Command word storage unit (third storage means)
DESCRIPTION OF SYMBOLS 109 Voice command input part 2 Cable relay member 201 Head mounting part 202 Connection part 203 Signal line 3 Adhesive member 4,4A Welding system 5 Welding power supply device 501 Power supply part 5011 Welding power supply 5012 Wire feeding power supply 502 Control part 503 Operation Unit 504 display unit 6,6A wire feeding device 601 wire feeding mechanism 602 gas solenoid valve 603 control unit 604 operation unit 605 display unit 606 voltage / current detection unit 607 wireless communication unit 607A antenna 7 welding torch 701 torch switch 8 gas supply Device 9 Wire reel 10 Remote operation device 1001 Battery 1002 Control unit 1003 Wireless communication unit 1003A Antenna 1004 Operation unit 1005 Display unit A Arc B Base material W Welding wire

Claims (8)

A voice command input device for inputting a command signal by voice,
Detecting means for detecting movement of the speaker's mouth when the speaker utters a voice command;
Command word conversion means for converting the movement of the mouth detected by the detection means into a command word uttered by the speaker;
Command signal generating means for generating the command signal corresponding to the command word converted by the converting means;
A voice command input device characterized by comprising: The detection means includes
An acceleration sensor attached to the periphery of the speaker's cheek or mouth;
First conversion means for converting a detection value of acceleration output from the acceleration sensor into data of movement of the mouth of the speaker in response to the speaker speaking a voice command;
Including
The command word conversion means includes
First storage means for storing a first conversion table indicating a correspondence between a phoneme uttered by the speaker and a mouth movement corresponding to the phoneme;
Second storage means for storing one or more command words registered in advance;
Second conversion means for converting the mouth movement data of the speaker output from the first conversion means into a phoneme string using the first conversion table;
Speech recognition means for recognizing a command word uttered by the speaker by collating a sequence of phonemes converted by the second conversion means with a command word registered in the second storage means;
Including
The command signal generating means includes
Third storage means for storing a second conversion table showing a correspondence relationship between one or more command words and the command signal stored in the second storage means;
Third conversion means for converting a command word recognized by the voice recognition means into a command signal corresponding to the command word using the second conversion table;
The voice command input device according to claim 1, comprising:   The speech recognition means calculates the degree of coincidence between the command word registered in the second storage means and the phoneme string converted by the second conversion means, and has a degree of coincidence equal to or greater than a predetermined threshold. The voice command input device according to claim 2, wherein a command word having the highest degree of coincidence is recognized as a voice command uttered by the speaker.   Mouth movement data according to the phoneme uttered by the speaker is displacement vector data of the acceleration sensor based on a change in facial expression when the speaker utters speech based on a facial expression that does not utter speech. The voice command input device according to claim 2 or 3.   5. The acceleration sensor according to claim 1, wherein the acceleration sensor is a triaxial acceleration sensor that detects accelerations in three axial directions orthogonal to each other, and is mounted on both cheeks or both sides of the mouth of the speaker. Voice command input device. Welding torch,
A wire feeding device for feeding a welding wire to the welding torch;
A welding power source for supplying welding power between the welding torch and the base material via the wire feeding device;
A command input device for inputting various commands to the welding power source;
A welding system comprising:
The said command input device is comprised by the audio | voice command input device in any one of Claims 1 thru | or 5, The welding system characterized by the above-mentioned.   The welding system according to claim 6, wherein the voice command input device is provided in the wire feeding device. A remote control device that can be connected to the wire feeding device by wireless communication,
The welding system according to claim 7, wherein the voice command input device is provided in the remote control device.

Images