JP7141938B2 - Voice recognition input device, voice recognition input program and medical imaging system - Google Patents

Description

TECHNICAL FIELD The present invention relates to a speech recognition input device, a speech recognition input program, and a medical imaging system, and more particularly, to a speech recognition input device, which is connected to a medical device such as a medical imaging device and outputs a command to the medical device. The present invention relates to a recognition input program and a medical imaging system.

In recent years, with the advancement of interventional treatment, there are an increasing number of cases in which blood vessels and gastrointestinal tracts are examined or treated under X-ray fluoroscopy and flexible endoscope manipulation. In these examinations or treatments, it is necessary to ensure the cleanliness of the operator, and manual operation of the equipment cannot maintain the cleanliness.
Further, for example, in an examination using an X-ray fluoroscope, equipment is often operated by a medical worker who assists the operator under verbal instructions of the operator who performs the examination procedure. In such a case, communication such as verbal instructions from the operator to the supporting medical staff takes time, and it may take time for the operator to operate the equipment as intended, and the operator is not assisted by the medical staff. It is desired that both medical devices are operated by direct verbal instructions.
Therefore, in medical equipment used for surgery, examination, etc., it is desired to operate by voice recognition in order to ensure cleanliness and improve operability.

On the other hand, in recent years, in addition to conventional methods using hidden Markov models, methods using deep learning have emerged in speech recognition technology. is improving. In addition, there are voice recognition processes that perform large-scale machine learning using servers and clouds to sequentially improve performance, but medical equipment must be designed with confidentiality in mind. A speech recognition input device applied to medical equipment needs to perform speech recognition processing in a non-network environment without connecting to Cloud or a server.

As an example of a medical device operated by voice recognition, Patent Document 1 discloses an X-ray image diagnostic apparatus that may harm a subject due to malfunction in order to reduce the burden on the operator. It is disclosed that the functions are controlled based on the manual operation by the operator, and the functions that are not likely to harm the subject even if they malfunction are controlled by recognizing the voice generated by the operator. .

JP 2006-149909 A

However, if, for example, a plurality of voice operation commands having similar phonemes are registered in the database used for voice recognition processing, there is a possibility of false detection occurring in voice recognition processing. That is, it is not possible to determine to which voice manipulation command the voice uttered by the operator corresponds to among a plurality of voice manipulation commands made up of similar phonemes, and there is a risk of erroneous detection. In this case, the operation cannot be performed by voice recognition processing, and as a result, the operator instructs the medical staff who provides support to operate the equipment, shortening the time required to operate the equipment as intended by the operator. I can't. Also, in that case, it is necessary to improve the recognition rate of voice operation commands, but the means for doing so is not specified.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and an object of the present invention is to accurately recognize operating instructions by voice of an operator and reduce erroneous detection in voice recognition processing.

In order to solve the above problems, the present invention provides the following means.
One aspect of the present invention is a voice recognition input device for inputting an operation command to an external device, in which a plurality of voice commands are associated with one operation command and recorded, and the voice command is used for each voice command. A storage unit storing a voice recognition data table in which weighting coefficients according to frequency are recorded, a voice input is received, voice recognition processing is performed using the voice input as a recognition target, and a voice command corresponding to the voice input is voice recognized. a command conversion unit for converting the voice command into an operation command recorded corresponding to the voice command by referring to the voice recognition data table; and converting the operation command to the external and an operation determination unit for outputting to a device, wherein the voice recognition unit selects a plurality of voice command candidates that can correspond to the voice input, and multiplies each of the plurality of voice command candidates by the weighting factor. To provide a voice recognition input device that outputs a voice command that seems to be a voice command as a result of voice recognition processing for the voice input.
According to the present invention, voice recognition processing is performed using a data table that records a weighting factor corresponding to the frequency of use for each voice command, so that the accuracy of voice recognition processing can be improved for operation instructions by voice.

According to the present invention, it is possible to reduce erroneous detection in voice recognition processing, and to accurately recognize operating instructions by the voice of the operator.

1 is a block diagram showing a schematic configuration of a speech recognition input device according to a first embodiment of the present invention; FIG. 2 is an example of a speech recognition data table stored in a speech recognition DB of the speech recognition input device of FIG. 1; 5 is a graph showing the relationship between the frequency of use of operation commands, operation command codes, and offset coefficients T, which are referred to when updating weighting factors in the voice recognition input device according to the first embodiment of the present invention. 4 is a flow chart showing the flow of speech recognition input processing by the speech recognition input device according to the first embodiment of the present invention; 9 is a graph showing the relationship between the cumulative number of voice commands, the voice command code, and the offset coefficient V, which are referred to when updating the weighting factor in the voice recognition input device according to the modification of the first embodiment of the present invention. 1 shows an example of a data table stored in a speech recognition DB in a speech recognition input device according to a second embodiment of the present invention, (A) is a data table showing the state of examination start information, (B) is an examination A data table showing the state of the type, (C) a data table showing the state of the X-ray irradiation information, and (D) a data table showing the device operation status and the classification of the speech recognition DB. 9 is a flow chart showing the flow of switching processing of the speech recognition data table in the speech recognition input device according to the second embodiment of the present invention;

A voice recognition input device according to an embodiment of the present invention performs input to an external device such as a medical imaging device connected to the voice recognition input device.

(First embodiment)
Hereinafter, the speech recognition input device according to the first embodiment of the present invention will be described in more detail with reference to the drawings. FIG. 1 shows a schematic configuration diagram of a speech recognition input device according to this embodiment.
The voice recognition input device 10 includes a central processing unit (CPU) 11 that controls the voice recognition input device 10 as a whole, a voice input I/F (interface) 12 that accepts voice input from a microphone or the like, a mouse, a keyboard, or the like. A manual input I / F (interface) 13 that accepts manual input (interface) 13, a memory 14, a voice recognition DB 15 that stores data necessary for voice recognition algorithms and voice recognition processing, and a log collection DB 16 that collects and records logs related to voice input. The configurations are connected together via a system bus.

In this embodiment, the voice recognition input device 10 is communicably connected to the imaging device 20 and performs various input instructions to the imaging device 20 . Further, the voice recognition input device 10 is connected to a display 30 via an imaging device 20, displays an image or the like acquired by the imaging device 20 on the display 30, and enlarges or reduces the displayed image. A desired operation instruction is given. Hardware for acquiring medical images, such as an X-ray device, an MRI device, a CT device, and a PET device, can be applied as the image capturing device 20 .

As shown in FIG. 1, the CPU 11 controls a voice operation processing unit 120, a manual operation processing unit 130, and a system operation determination unit 140 in order to issue voice input instructions to an imaging device or the like using the voice recognition input device 10. Realize the function. In particular, the voice operation processing unit 120 implements the functions of the voice recognition unit 111 , the command conversion unit 112 and the command analysis unit 113 .

The functions of these units realized by the CPU 11 can be realized as software by the CPU reading in advance a program stored in a memory such as a magnetic disk (not shown) and executing the program. Part or all of the operations performed by each unit included in the CPU 11 can also be realized by ASIC (application specific integrated circuit) or FPGA (field-programmable gate array).

The voice operation processing unit 120 recognizes a voice operation instruction (voice input) input via the voice input I/F 12 such as a microphone, and issues an operation instruction to the imaging device 20. It implements the functions of the recognition unit 111 , the command conversion unit 112 and the command analysis unit 113 .

The speech recognition unit 111 performs speech recognition processing on a speech operation instruction input via the speech input I/F 12 according to a speech recognition algorithm stored in advance in the speech recognition DB 15 or the like, and recognizes a speech command as a recognition result. to the command conversion unit 112 . Here, in the speech recognition process, the speech recognition unit 111 uses a speech recognition data table (see FIG. 2) stored in the speech recognition DB 15, which will be described later, to perform speech recognition processing according to a predetermined speech recognition algorithm. Pick a command. Details of the speech recognition processing by the speech recognition unit 111 will be described later.

The command conversion unit 112 converts the voice command selected through the voice recognition processing in the voice recognition unit 111 into an operation command corresponding to the voice command, and converts the operation command into an operation instruction by the operator, which is used by the system operation determination unit 140 and the command analysis unit 113. output to

The command analysis unit 113 acquires information about operation commands related to operation instructions by the operator from the command conversion unit 112, generates an operation history, records it in the log collection DB 16, and analyzes the operation history. Here, the information about the operation command can include not only the operation command but also the voice command before being converted into the operation command. Also, the command analysis unit 113 updates the weighting factor of the voice recognition data table based on the analysis result of the operation history. The details of updating the weighting factors will be described later.

When a manual operation is performed, the manual operation processing unit 130 generates an operation command based on an operation instruction input via the manual input I/F 13 such as a mouse and a keyboard, and performs system operation. Output to determination unit 140 .
The system operation determination unit 140 outputs the operation command input from the voice operation processing unit 120 or the manual operation processing unit 130 to the imaging device 20 and to the command analysis unit 113 .

The voice input I/F 12 receives speech of an operator such as an operator as a voice operation instruction (voice input), converts it into voice data of an electric signal, and outputs the voice data to the voice operation processing unit 120 . , for example a microphone or the like can be applied.
The manual input I/F 13 receives a manual operation instruction from an operator or the like, converts the received operation instruction into an electric signal, and outputs the electric signal to the manual operation processing unit 130. For example, a mouse, a keyboard, a touch panel, or the like. Input devices can be applied.

The memory 14 stores the programs executed by the CPU 11 and intermediate progress of the arithmetic processing, and temporarily stores voice and manual operation instructions.
The speech recognition DB 15 stores a predetermined speech recognition algorithm and records a speech recognition data table used for speech recognition processing. Details of the voice recognition data table will be described later.
The log collection DB 16 acquires and records information related to operation instructions such as operation history generated by the command analysis unit 113 .

(Regarding voice recognition processing and voice recognition data table)
As shown in FIG. 2, the voice recognition data table includes voice command data db1 indicating data relating to voice commands, operation command data db2 indicating data relating to operation commands, and weighting factors indicating weighting factors determined for each voice command. It consists of data db3.

As shown in FIG. 2, in the voice recognition data table, a plurality of voice commands db11 are recorded in association with the same operation command db21. In this way, even when the same operation instruction is given by different operators using different phrases or utterances, the same operation can be realized through voice recognition processing. Also, a command reading db12 and a voice command code db13 are recorded in association with each voice command of the voice command data db1. As for the operation command db21, similarly, an operation command code db22 is associated with each operation command and stored.

By the way, the speech recognition unit 111 performs speech recognition processing as follows. That is, the speech recognition unit 111 first converts speech data input via the speech input I/F into sound waves, decomposes each character of the speech data from the sound waves into phonemes, and specifies the phonemes. Subsequently, using a speech recognition algorithm such as statistical machine learning based on a hidden Markov model or machine learning using a deep learning model, the phonemes of the speech data are compared with the phonemes of the command reading db12. By this collation, the voice recognition unit 111 selects command readings db12 similar to the voice data, and outputs voice command db11 candidates corresponding to the selected command readings and scores as indicators of their likelihood.

Here, as an example of voice recognition processing, a case in which the operator utters "reduce image" to give an operation instruction will be considered. When the operator utters "reduce image", the phonemes of "gazoshukusho" and "gazoshushuu" match halfway in the command reading db12 of the speech recognition data table. For this reason, the speech recognition unit 11 may mistakenly recognize "collection of images" as "collection of images" depending on the quality of the input speech data. In this case, even though the operator instructed image reduction of the operation command db21, the operation command db21 "perspective record" corresponding to the voice command db11 "remember" is selected as the operation command due to misrecognition of the voice recognition. and an operation unintended by the operator is performed.

Therefore, in order to avoid such erroneous recognition, the voice recognition data table records the weighting factor data db3 in association with each voice command. The weighting factor data db3 is a weighting factor to be multiplied by the score attached to each of the one or more voice command candidates output in the process of speech recognition processing. The weighting factor recorded in the weighting factor data db3 is a value that corresponds to each voice command and is determined according to the frequency of use of the voice command.

Also, the weighting factor can be updated based on the result of analysis of the operation history recorded in the log collection DB 16 by the command analysis unit 113 . That is, the command analysis unit 113 analyzes the operation history to calculate the number of times each operation command is used within a certain period of time, and calculates the offset coefficient T for updating the weighting factor based on the number of times of use.

The offset coefficient T can be determined, for example, by the accumulated number of times the operation command code corresponding to each operation command is issued according to the graph shown in FIG. In addition, the offset coefficient T can also be calculated based on the ratio of each operation command to the total number of issued operation command codes in a predetermined period. The command analysis unit 113 multiplies each weighting factor recorded corresponding to the operation command by the calculated offset factor T to update the weighting factor.

The analysis of the operation history by the command analysis unit 113 can be performed automatically, or can be performed at arbitrary timing, such as in accordance with an operation instruction from the operator or the device provider. Further, when an erroneous detection occurs in the voice recognition processing of the voice recognition unit 111, the command analysis unit 113 can update the weighting factor by subtracting it from the number of times the operation command is used.

The flow of speech input processing by the speech recognition input device 10 configured as described above will be described below with reference to the flowchart of FIG.
As shown in FIG. 4, when the voice recognition input device 10 is activated, voice input is put into a standby state. When there is a voice input in the voice input I/F (step S101), the process proceeds to step S102, and the voice recognition unit 111 receives input of voice data from the voice input I/F 12 and recognizes the phonemes of the voice data and the voice recognition DB 15. , and selects voice command candidates recorded corresponding to command readings. Selection of this candidate can be determined based on the score, which is an index of certainty of speech recognition.

In the next step S103, it is determined whether or not there is one or more voice command candidates selected in step S102. back to If the number of voice command candidates is one or more, the process advances to step S105 to calculate the score assigned to each voice command candidate and the weighting factor recorded corresponding to the voice command in the voice recognition data table. Multiply.

In step S106, the voice command candidate with the highest score is selected as a result of the multiplication of the score and the weighting factor in step S105. In the next step S107, it is determined whether the highest score, that is, the score of the selected voice command candidate is greater than a predetermined threshold. If the highest score is less than the predetermined threshold, there is no voice command. After passing through step S104, the process returns to step S101 without executing the voice operation, and the voice recognition input device 10 again waits for voice input. At this time, it is possible to notify the operator that the voice operation will not be performed by using, for example, a synthesized voice, an alarm, or a display.

If the highest score is greater than the predetermined threshold, the process proceeds to step S108 to determine the voice command indicating the highest score as the result of voice recognition processing. The determined voice command is output to the command conversion unit 112, and the command conversion unit 112 uses the voice recognition data table to convert the determined voice command into an operation command corresponding to the voice command (S109).

In the next step S<b>110 , the command conversion section 112 outputs the converted operation command to the command analysis section 113 and the system operation determination section 140 . The command analysis unit 113 updates and generates an operation history including the input operation command, and records it in the log collection DB 16 . The system operation determination unit 140 outputs the input operation command to the imaging device 20 . The imaging device 20 executes an operation according to the input operation command.

As described above, according to the present embodiment, a plurality of voice commands are associated with an operation command and recorded, and each operation command is given a weighting factor in descending order of frequency of use, and voice recognition processing is performed. It is possible to accurately perform speech recognition processing without depending on the utterance habits and preferences that differ from operator to operator. In addition, by recording the frequency of use of the operation command and updating the weighting factor according to the frequency of use, it is possible to improve the accuracy of voice recognition processing for the operation command that is frequently used by the operator over time. It is possible to accurately recognize operation instructions by the voice of the operator.

(Modification)
In the first embodiment described above, an example has been described in which the command analysis unit 113 updates the weighting factor based on the frequency of use of the operation command. In this modified example, an example of updating weighting factors based on an input voice command will be described.

The command analysis unit 113 analyzes the operation history to calculate the frequency of detection of each voice command within a certain period of time before being converted into an operation command input from the command conversion unit 112 . Then, an offset coefficient V for updating the weighting coefficient is calculated based on the calculated voice command detection frequency.

The offset coefficient V can be determined, for example, according to the graph shown in FIG. 5, based on the accumulated number of times the voice command code corresponding to each voice command has been issued. Alternatively, the offset coefficient V may be determined based on the ratio of each voice command to the total number of voice command codes issued during a certain period.

The command analysis unit 113 multiplies each weighting factor recorded corresponding to the operation command by the calculated offset factor V to update the weighting factor. In this case as well, when an erroneous detection occurs in the voice recognition processing of the voice recognition unit 111, the command analysis unit 113 preferably updates the weighting factor by subtracting it from the cumulative number of voice commands.

As described above, in this modified example, a higher weighting factor is given to voice commands based on the operator's utterances in descending order of detection frequency. It is possible to accurately perform speech recognition processing without depending on In addition, by recording the number of voice command detections and updating the weighting factor according to the number of detections, it is possible to improve the accuracy of voice recognition over time for voice commands that are frequently detected by the operator. The accuracy of speech recognition can be further improved as compared with the above-described first embodiment, and the operator's voice operation instructions can be accurately recognized.

It is also possible to update by multiplying the result of multiplying both the offset coefficient T based on the frequency of use of the operation command and the offset coefficient V based on the detection frequency of the voice command by the weighting coefficient. In this case, the accuracy of voice recognition for frequently used operation commands and frequently detected voice commands is further improved. In this case as well, when an erroneous detection occurs in the voice recognition processing of the voice recognition unit 111, the command analysis unit 113 subtracts the number of times the operation command is used and the cumulative number of times the voice command is accumulated, and sets the weighting factor. preferably updated.

(Second embodiment)
In the first embodiment and its modified example described above, an example in which one voice recognition data table is stored in the voice recognition DB 15 has been described. In this embodiment, the voice recognition DB 15 stores a plurality of voice recognition data tables as shown in FIG. The data table used for speech recognition processing is switched according to the operational status.

As the operation status of the apparatus, for example, before and after the start of examination, examination type, presence/absence of X-ray output, etc. can be considered. store. An example of the voice recognition data table is as shown in FIG. Further, the speech recognition DB stores data tables showing the operational status of the apparatus as shown in FIG. 6, and an appropriate speech recognition data table is selected by referring to these data tables. 6(A) is a data table showing the state of examination start information, (B) is a data table showing the state of examination types, (C) is a data table showing the state of X-ray irradiation information, (D ) is a data table showing the device operation status and the classification of the speech recognition database.

The switching flow of the speech recognition data tables tg001 to tg*** will be described below with reference to the flowchart of FIG.
When the voice recognition input device 10 is activated, the system operation determination unit 14 sequentially acquires information on the device operation status from the imaging device 20 (step S201). In this embodiment, the system operation determination unit 14 acquires, for example, examination start information, examination type information, and X-ray irradiation information. After obtaining the information about the device operation status, the system operation determining unit 14 determines whether or not each of the obtained information has changed from the previous state (step S202).

If it is determined in step S202 that at least one of the examination start information, examination type information, and X-ray irradiation information has changed, the process advances to step S203 to generate an apparatus operation status command St. As shown in FIG. 6D, the apparatus operation status command St consists of three pieces of information: examination start information, examination type, and X-ray irradiation information. It's like

The system operation determination unit 14 outputs the generated device operation status command St to the voice recognition unit 111 (step S205), and the voice recognition unit 111 selects a voice recognition data table according to the input device operation status command St. , switch. Each information of the apparatus operation status command St, for example, examination start information ao1 indicating "after examination start", examination type 1001 indicating "abdomen (abdomen)", X-ray irradiation information c01 indicating "under irradiation" In some cases, the speech recognition database tg003 is selected.

Even when a plurality of speech recognition data tables are stored in the speech recognition DB as described above, weighting factors can be updated in the same manner as in the first embodiment and its modification.

When generating the operation history, the command analysis unit 113 records the command code (see FIG. 6) related to whether or not to start the examination and the examination type, together with the operation command code indicating the operation command and the voice command code indicating the voice command. . By doing so, it is possible to update only the weighting coefficients of the speech recognition database that need to be updated.

As described above, according to the present embodiment, it is possible to switch the speech recognition data table used for the speech recognition process according to the operational status of the device to which the speech recognition input device is applied. Each voice recognition data table can define a weighting factor that places a weight on frequently used operations and frequently used voice commands for each device operation status, thereby improving the accuracy of voice recognition processing.

DESCRIPTION OF SYMBOLS 10... Voice recognition input device, 11... CPU, 12... Voice input I/F, 13... Manual input I/F, 14... Memory, 15... Voice recognition DB, 16 ... log collection DB, 20 ... image pickup device, 30 ... display, 111 ... speech recognition section, 112 ... command conversion section, 113 ... command analysis section, 120 ... voice Operation processing unit 130 Manual operation processing unit 140 System operation determination unit 20 Image pickup device 30 Display

Claims (6)

A voice recognition input device for inputting an operation command to an external device,
a storage unit storing a voice recognition data table in which a plurality of voice commands are associated with one operation command and recorded, and a weighting factor corresponding to the frequency of use of the voice command is recorded for each voice command;
a voice recognition unit that receives a voice input, performs voice recognition processing on the voice input as a recognition target, and outputs a voice command corresponding to the voice input as a result of the voice recognition processing;
a command conversion unit that refers to the voice recognition data table and converts the voice command into an operation command recorded in correspondence with the voice command;
an operation determination unit that outputs the operation command to the external device;
The speech recognition unit refers to the speech recognition data table, selects a plurality of speech command candidates that can correspond to the speech input, and calculates an index of likelihood of speech recognition, and for each of these plurality of speech command candidates : A voice recognition input device for selecting the most probable voice command by multiplying the index and the weighting factor, and outputting the selected voice command as a result of voice recognition processing for the voice input. A command analysis unit that generates an operation history recording at least one of the operation command and the voice command corresponding to the operation command, and updates the weighting factor based on a result of analyzing the operation history. 1. The speech recognition input device according to 1. 3. The voice recognition input device according to claim 2, wherein the command analysis unit updates the weighting factor based on at least one of the frequency of use of the operation command within a certain period of time and the cumulative number of voice commands. the storage unit stores a plurality of voice recognition data tables corresponding to the operation status of the external device;
5. The speech recognition input device according to claim 1, wherein the speech recognition unit switches the speech recognition data table used for speech recognition processing according to the operational status of the external device. A voice recognition input program that causes a computer to input an operation command by voice to a medical imaging device,
A plurality of voice commands are associated with one operation command and recorded, and a voice input from the user is recognized by referring to a voice recognition data table in which a weighting factor corresponding to the frequency of use of each voice command is recorded. a speech recognition step of performing speech recognition processing as a recognition target and outputting a speech command corresponding to the speech input as a result of the speech recognition processing;
a command conversion step of referring to the voice recognition data table and converting the voice command into an operation command recorded corresponding to the voice command;
an operation decision step of outputting the operation command to the medical imaging device;
In the voice recognition step, referring to the voice recognition data table, a plurality of voice command candidates that can correspond to the voice input are selected and an index of likelihood of voice recognition is calculated, and for each of the plurality of voice command candidates , A voice recognition input program for selecting the most probable voice command by multiplying the index and the weighting factor, and outputting the selected voice command as a result of voice recognition processing for the voice input. a speech recognition input device according to any one of claims 1 to 4;
5. A medical device, comprising a medical imaging device as an external device, wherein the voice recognition input device according to any one of claims 1 to 4 issues operation instructions to the medical imaging device by voice recognition input. Image capture system.

Images