Word registration method and word registration device

Abstract

To provide a word registration method and a word registration device that make it efficient to register words in a pronunciation dictionary of a speech recognition engine used to steer a moving body.SOLUTION: A word registration method is for registering words in a pronunciation dictionary of a speech recognition engine which inputs a speech pronounced in steering a moving body, and includes: a step of extracting words included in the speech pronounced in steering the moving body; a first integration step of combining, when the extracted words include a plurality of words having the same array pattern of vowels, one of the plurality of words having the same array pattern of vowels with words before and after the one in the speech so as to be integrated into one word; and a step of registering the extracted words and the words integrated into one word in the pronunciation dictionary.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a word registration method and a word registration device. [Background technology]

[0002] Normally, in a moving object such as a ship, the pilot listens to the steering commands issued by the commander (such as the captain), understands their meaning, and steers the ship. In recent years, in order to promote manpower saving due to the decrease in the workforce, it is desirable to use a voice recognition engine to steer the ship by the commander's voice command without the need for a pilot. To achieve this, it is necessary to improve the accuracy of voice recognition of steering commands.

[0003] As a technique related to the present disclosure, Patent Document 1 discloses a technique for identifying instruction action information by performing a voice recognition process on instruction voice data. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-214466 A Summary of the Invention [Problem to be solved by the invention]

[0005] In a typical speech recognition engine, you can customize the words to be recognized as a single word by registering a combination of words (text information) and pronunciations (acoustic models) in a pronunciation dictionary. In order to improve the recognition rate of a speech recognition engine, it is necessary to create this pronunciation dictionary, but this is time-consuming because it requires trial and error while understanding the tendency of the speech recognition engine to make incorrect recognitions.

[0006] An object of the present disclosure is to improve the efficiency of the process of registering words in a pronunciation dictionary of a voice recognition engine used for operating (steering) a moving object. [Means for solving the problem]

[0007] According to one aspect of the present disclosure, a word registration method is a method for registering words in a pronunciation dictionary of a voice recognition engine used to operate a moving body, and includes the steps of: extracting words included in lines spoken during the operation; if there are multiple extracted words with the same vowel arrangement pattern, performing a first combination process of combining one of the words with the same vowel arrangement pattern with the words before and after it in the line to combine them into a single word; and registering the extracted words and the words combined into a single word in the pronunciation dictionary.

[0008] According to one aspect of the present disclosure, a word registration device is a device that registers words in a pronunciation dictionary of a voice recognition engine used to operate a moving body, and includes an extraction unit that extracts words included in lines spoken during the operation, a first combination processing unit that, when there are multiple extracted words with the same vowel arrangement pattern, combines one of the words with the same vowel arrangement pattern with the words before and after it in the line to combine them into a single word, and a registration unit that registers the extracted word and the words combined into a single word in the pronunciation dictionary.

[0009] According to one aspect of the present disclosure, a word registration method is a method for registering words in a pronunciation dictionary of a voice recognition engine used to operate a moving body, and includes the steps of: extracting words contained in lines spoken during the operation; combining two words belonging to a pair of words used in combination in the line into one word when the final vowel of the first word matches the first vowel of the second word; and registering the extracted words and the words combined into one word in the pronunciation dictionary. Effect of the Invention

[0010] According to each of the above-described aspects, it is possible to improve the efficiency of the process of registering words in the pronunciation dictionary of a voice recognition engine used to operate a moving object. [Brief description of the drawings]

[0011] [Figure 1] 1 is a diagram showing the overall configuration of a steering control device according to a first embodiment. [Diagram 2] FIG. 2 is a diagram showing an example of a data structure of a pronunciation dictionary according to the first embodiment. [Diagram 3] FIG. 4 is a diagram showing a flow of a process for registering words in a pronunciation dictionary according to the first embodiment. [Figure 4] 5A to 5C are diagrams illustrating specific examples of each process in a registration operation according to the first embodiment. [Diagram 5] 5A to 5C are diagrams illustrating specific examples of each process in a registration operation according to the first embodiment. [Figure 6] FIG. 11 is a diagram illustrating a configuration of a word registration device according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] <First embodiment> Hereinafter, a method for registering words in a pronunciation dictionary according to the first embodiment will be described in detail with reference to FIGS.

[0013] (Overall configuration of the flight control device) FIG. 1 is a diagram showing the overall configuration of a steering control device according to a first embodiment. The operation control device 9 shown in Fig. 1 is used for steering a ship, which is an example of a moving body. This operation control device automatically (without the intervention of a pilot) controls the operation (steering) of the ship in accordance with a steering command issued by a person issuing the command.

[0014] The operation control device 9 includes a voice recognition engine 1 for recognizing spoken steering commands. The voice recognition engine 1 inputs the spoken voice of the steering command and converts it into text information. The operation control device 9 interprets the meaning of the spoken content (steering command) based on the text information converted by the voice recognition engine 1, and performs steering control. The voice recognition engine 1 according to this embodiment may be one that utilizes existing technology.

[0015] The voice recognition engine 1 includes a pronunciation dictionary 10. The pronunciation dictionary 10 is an information table in which words to be recognized (text information) and their pronunciations (acoustic models) are registered in association with each other. By referring to the pronunciation dictionary 10, the voice recognition engine 1 identifies words that match the input voice with the acoustic model, and converts the words into text information.

[0016] (Example of a pronunciation dictionary) FIG. 2 is a diagram illustrating an example of a data structure of the pronunciation dictionary according to the first embodiment. As shown in FIG. 2, the pronunciation dictionary 10 associates words (text) with pronunciations (acoustic models). The "words" in Figure 2 are text information, and in addition to Arabic numerals such as "2" and "3," words such as "starboard" and "full" are registered. The "pronunciations" in Figure 2 are the pronunciations that correspond to each "word." When a word has multiple pronunciations, all of them are matched. For example, the number "2" is associated with multiple pronunciations such as "NI," "NII," and "HUTA."

[0017] (Pronunciation dictionary registration process) FIG. 3 is a diagram showing a flow of a process for registering words in a pronunciation dictionary according to the first embodiment. 4 and 5 are diagrams showing specific examples of each process in the registration operation according to the first embodiment. The flow of the word registration method according to this embodiment will be described in detail below with reference to FIGS.

[0018] The process flow shown in FIG. 3 shows the flow of operations for registering words to be recognized by the voice recognition engine 1 in the pronunciation dictionary 10 at a stage before the operation of the operation control device 9 starts.

[0019] As shown in FIG. 3, first, the worker overviews all the lines used as steering commands for ships (e.g., about 100 types of lines) and extracts words to be registered (words to be registered as candidates) (step S01). Here, for example, for frequently used lines such as "torikajiippai" (full rudder) and "omokajiippai" (full rudder), the lines themselves are added as individual words to the registration candidates. In addition, since "ippai" (full) is also used in other lines, it is added to the registration candidates as a separate word from "torikajiippai" and "omokajiippai" (full). In addition, as described above, for words that have multiple readings, such as "2" (ni, ni, futa), all of the pronunciations are associated and added to the registration candidates. Hereinafter, the list of all combinations of words and pronunciations listed as registration candidates in step S01 will also be referred to as a "group of registration candidate words."

[0020] Next, the worker determines whether or not there are multiple words with the same vowel sequence pattern among the group of registration candidate words listed in step S01 (step S02). If there are multiple words with the same vowel sequence pattern (step S02; YES), the worker replaces one of the words with another word combined with the words before and after it in the dialogue (step S03).

[0021] Here, "vowel arrangement pattern" refers to the types of vowels (five types: "A," "I," "U," "E," and "O"), as well as their number and order in the pronunciation of each word. "Matching vowel arrangement patterns" refers to the complete match of the types, number, and order of vowels in the pronunciation of each word.

[0022] The processes in steps S02 and S03 will be described in more detail with reference to FIG.

[0023] As shown in Figure 4, suppose that the group of candidate words for registration includes the word "8" (HACHI) and the word "KAJI" (KAJI). In this case, these two words "8" and "KAJI" have the same vowel sequence pattern of "A" → "I".

[0024] In this case, the operator selects one of the target words (either "8" or "rudder") and replaces it with a single word that combines the word with the words before and after it in the dialogue. For example, if the operator selects "rudder," the operator checks the words before and after "rudder" in all the dialogues used for steering commands.

[0025] As a result, suppose that the word "rudder" is used in two ways in all lines of steering commands: "turn the rudder" and "take the rudder." In this case, the worker removes the word "rudder" (KAJI) from the group of candidate words to be registered, and instead adds the two words "turn the rudder" (KAJIKIRE) and "take the rudder" (KAJITORIRE) to the list of candidates to be registered. Here, the word "rudder" is a single word formed by combining "rudder" and "kire" (cut) that are in a contextual relationship in the lines, and the word "take the rudder" is a single word formed by combining "rudder" and "take" (take) that are in a contextual relationship in the lines. In this way, the word "rudder" is replaced with the two words "rudder cut" and "rudder take". Note that the two newly added words "rudder cut" and "rudder take" do not have the same vowel arrangement pattern.

[0026] Returning to Fig. 3, the worker next determines whether or not there is a pair of two words used in a dialogue in which the final vowel of the first word matches the first vowel of the second word (step S04) for the group of candidate words for registration obtained through steps S01 to S03. If there is a pair of words in which the final vowel of the first word matches the first vowel of the second word (step S04; YES), the worker adds the two words together to the group of candidate words for registration (step S05).

[0027] The processes in steps S04 and S05 will be described in more detail with reference to FIG.

[0028] As shown in FIG. 5, suppose that the group of candidate words for registration includes the word "speed" (HAYASA) and the word "three" (SAN). These two words, "speed" and "three," may be used one after the other in a line. For example, the line "Speed ​​30" is composed of the word "speed" (HAYASA) and the word "three" (SAN) connected in that order. Furthermore, the final vowel "A" of the first word "speed" (HAYASA) matches the first vowel "A" of the second word "three" (SAN).

[0029] In this case, the worker combines the pair of words "speed" and "3" into a single word "speed 3" (HAYASASAN), and adds this to the group of candidate words to be registered.

[0030] Similarly, suppose the group of registered candidate words includes the word "5" (GO) and the word "degree" (DO). These two words, "5" and "degree", may be used one after the other in a line. For example, the line "15 degrees" is composed of the word "5" (GO) and the word "degree" (DO) connected in that order. Furthermore, the final vowel "O" of the first word "5" (GO) matches the first vowel "O" of the second word "degree" (DO).

[0031] In this case, the worker combines the pair of words "5" and "degree" into a single word "5 degrees" (GODO), and adds this to the group of candidate words to be registered.

[0032] 3, next, the worker registers the words included in the group of registration candidate words obtained through the processes of steps S02 to S03 and steps S04 to S05 in the pronunciation dictionary, and then checks the actual recognition rate for each word (step S06). It is determined whether the recognition rate for each word is less than 100%, and for words that are less than 100% (step S07; NO), the pronunciation dictionary is individually customized for each word (step S08). Examples of individual customization for each word are as follows. (Example) In response to a misrecognition in which a number not found in the dictionary dialogue (steering commands) appears (for example, the dictionary dialogue "hard starboard" is misrecognized as "hard starboard 5"), logic processing is applied to limit the use of numbers to only be used with units such as "degrees".

[0033] When the recognition rate for all words is 100%, the registration process ends (step S07; YES).

[0034] (Action, effect) As described above, the method for registering the pronunciation dictionary according to this embodiment is characterized in that, when there are multiple words with the same vowel arrangement pattern in a group of candidate words for registration, a process (first combining process) is performed in which one of the words with the same vowel arrangement pattern is combined with the words before and after it in the dialogue to form a single word (steps S02 to S03 in FIG. 3).

[0035] Here, we investigated the tendency of misrecognition by the voice recognition engine 1 (a voice recognition engine based on existing technology), and found that words with the same vowel arrangement pattern tend to be easily misrecognized by each other. For example, we found that the word "8" (HACHI) and the word "rudder" (KAJI) are easily misrecognized by the voice recognition engine. By doing as described above, it is possible to avoid registering words that are easily misrecognized with each other in the pronunciation dictionary. For example, the word "8" (HACHI) and the word "rudder" (KAJI) are easily misrecognized, but the relationship between "8" (HACHI) and "KAJIKIRE" (KAJIKIRE) or "KAJITORE" (KAJITORE) is not likely to be misrecognized by the speech recognition engine 1 because the vowel arrangement patterns are different.

[0036] Furthermore, the method of registering a pronunciation dictionary according to this embodiment is characterized in that, when the final vowel of the first word of a pair of two words used together in dialogue matches the first vowel of the second word, the method performs a process of combining the two words belonging to the pair into one word (a second combining process) (steps S04 to S05 in FIG. 3).

[0037] Here, as a result of investigating the tendency of misrecognition by the speech recognition engine 1, it was found that when the final vowel of the first word of a pair of two consecutive words matches the first vowel of the second word, misrecognition is likely to occur in the speech recognition engine 1. For example, when "speed 3" is included in a line, the speech recognition engine 1 tends to have difficulty recognizing this as a combination of the words "speed" and "3", and similarly, when "fifth" is included in a line, the speech recognition engine 1 tends to have difficulty recognizing this as a combination of the words "5" and "degree". Therefore, by doing as described above, the phrase "speed 3" that appears in the dialogue is recognized as a single word, "speed 3" (HAYSASAN). This allows the voice recognition engine 1 to recognize the phrase "speed 3" as a single word, rather than as a combination of the two words "speed" and "3", thereby reducing misrecognition.

[0038] Moreover, in the pronunciation dictionary registration method according to this embodiment, steps S02 to S03 (first combining process) in FIG. 3 are performed, and then steps S04 to S05 (second combining process) are performed. In this way, by first executing the processes in steps S02 and S03, it is possible to reduce the number of words to be processed in steps S04 and S05, and therefore the overall processing load of the process flow in FIG.

[0039] (Modification of the first embodiment) In the above-mentioned first embodiment, both steps S02 to S03 (first combining process) and steps S04 to S05 (second combining process) in Fig. 3 are described as being included, but other embodiments are not limited to this aspect. For example, in the word registration method according to other embodiments, only one of steps S02 to S03 (first combining process) and steps S04 to S05 (second combining process) may be performed.

[0040] In the above-mentioned first embodiment, the words are replaced in step S03 (first combining process) in Fig. 3, but other embodiments are not limited to this. For example, in other embodiments, the words combined into one word in step S03 in Fig. 3 may be added to the registration candidate word group.

[0041] <Second embodiment> In the first embodiment, the registration process is performed manually by an operator through the process flow shown in Fig. 3, but other embodiments are not limited to this. That is, the process flow shown in Fig. 3 may be automatically performed by a device. The word registration device according to the second embodiment will be described with reference to FIG.

[0042] (Configuration of the word registration device) FIG. 6 is a diagram showing a configuration of a word registration device according to the second embodiment. 6, the word registration device 2 is a device that registers words in a pronunciation dictionary 10 of a voice recognition engine 1 used for steering (operating) a ship (mobile body). The word registration device 2 includes a CPU 20, a memory 21, an input / output interface 22, and a recording medium 23 as a general computer configuration.

[0043] The CPU 20 is a processor that controls the overall operation of the word registration device 2, and performs various functions by operating according to a predetermined program. Each function of the CPU 20 will be described later. The memory 21 is a so-called main storage device, into which programs for operating the CPU 20 and the like are loaded. The input / output interface 22 is a communication interface for exchanging information with external devices. In this embodiment, the input / output interface 22 receives input of information indicating the contents of all lines used as steering commands. The recording medium 23 is a large-capacity storage device such as an HDD or SSD.

[0044] The CPU 100 operates according to a program to function as an extraction unit 201, a first combination processing unit 202, a second combination processing unit 203, and a registration unit 204.

[0045] The extraction unit 201 extracts words contained in lines uttered during steering (operation) (step S01 in FIG. 3).

[0046] When there are multiple extracted words with the same vowel sequence pattern, the first combination processing unit 202 combines one of the words with the same vowel sequence pattern with the words before and after it in the dialogue to create one word (steps S02 to S03 in FIG. 3).

[0047] The second combining processor 203 combines two words that are used together in a line into one word when the final vowel of the first word matches the initial vowel of the second word (steps S04 to S05 in FIG. 3).

[0048] The registration unit 204 registers the extracted words and the words grouped into one word in the pronunciation dictionary 10 (step S06 in FIG. 3).

[0049] In the above-mentioned embodiment, the various processes executed by the word registration device 2 are stored in a computer-readable recording medium in the form of a program, and the above-mentioned various processes are executed by the computer reading and executing the program. The computer-readable recording medium refers to a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, etc. The computer program may be distributed to a computer via a communication line, and the computer that receives the program may execute the program.

[0050] The program may be for implementing some of the above-mentioned functions, or may be a so-called differential file (differential program) that can implement the above-mentioned functions in combination with a program already recorded in the computer system.

[0051] As described above, several embodiments according to the present disclosure have been described, but all of these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents as described in the claims, as well as in the scope and gist of the invention.

[0052] <Additional Notes> The word registration method and the word registration device described in each embodiment can be understood, for example, as follows.

[0053] (1) In a first aspect, a word registration method is a method for registering words in a pronunciation dictionary 10 of a voice recognition engine 1 used for steering (piloting) a ship (mobile body), and includes the steps of: extracting words included in lines uttered during steering; if there are multiple extracted words with the same vowel arrangement pattern, performing a first combination process of combining one of the words with the same vowel arrangement pattern with the words before and after it in the lines to combine them into one word; and registering the extracted words and the words combined into one word in the pronunciation dictionary.

[0054] (2) In a second aspect, the word registration method further includes a step of performing a second combining process of combining two words belonging to a pair of words used together in dialogue into one word when the final vowel of the first word matches the initial vowel of the second word.

[0055] (3) In a third aspect, the word registration method includes executing a first combining process and then executing a second combining process.

[0056] (4) In a fourth aspect, the word registration device 2 is a device for registering words in a pronunciation dictionary of a voice recognition engine used for operating a moving body, and includes an extraction unit 201 for extracting words included in lines spoken during operation, a first combination processing unit 202 for combining one of the words having the same vowel arrangement pattern with the words before and after it in the line to form a single word when there are multiple extracted words having the same vowel arrangement pattern, and a registration unit 204 for registering the extracted word and the words combined into a single word in the pronunciation dictionary.

[0057] (5) In a fifth aspect, a word registration method is a method for registering words in a pronunciation dictionary of a voice recognition engine used for operating a moving body, comprising the steps of: extracting words contained in lines spoken during the operation; combining two words that are used in combination in the lines and in which the final vowel of the first word matches the first vowel of the second word; and registering the extracted words and the combined words in the pronunciation dictionary. [Explanation of symbols]

[0058] 9. Steering control device 1. Speech recognition engine 10 Pronunciation Dictionary 2. Word registration device

Claims

1. A method for registering words into a pronunciation dictionary of a voice recognition engine used for operating a moving object, comprising the steps of: A step of extracting words included in lines uttered during the operation; a step of performing a first combining process in which, when there are a plurality of words having the same vowel arrangement pattern among the extracted words, one of the words having the same vowel arrangement pattern is combined with a word before and after the word in the dialogue to form a single word; registering the extracted words and the words grouped into one word in the pronunciation dictionary; The word registration method includes the steps of:

2. The method further includes a step of performing a second combining process for combining two words belonging to a pair of words used in conjunction in the dialogue into one word when a final vowel of a first word and an initial vowel of a second word match each other. The word registration method according to claim 1 .

3. performing the second combining process after performing the first combining process; The word registration method according to claim 2.

4. An apparatus for registering words in a pronunciation dictionary of a voice recognition engine used for operating a mobile object, comprising: An extraction unit that extracts words included in lines uttered during the operation; a first combining unit that combines, when there are a plurality of words having the same vowel arrangement pattern among the extracted words, one of the words having the same vowel arrangement pattern with a word before and after the word in the dialogue to form a single word; a registration unit that registers the extracted words and words grouped into one word in the pronunciation dictionary; A word registration device comprising:

5. A method for registering words into a pronunciation dictionary of a voice recognition engine used for operating a moving object, comprising the steps of: A step of extracting words included in lines uttered during the operation; a step of combining two words belonging to a pair of words used in conjunction in the dialogue into one word when a final vowel of a first word and an initial vowel of a second word are the same; registering the extracted words and the words grouped into one word in the pronunciation dictionary; The word registration method includes the steps of:

Images