Language processing system, language processing method, and language processing program

Abstract

To provide a technique to obtain more accurate results for words that have multiple readings.SOLUTION: In a language processing system, a text is configured to be passed between processes. Voice recognition process is configured to output corresponding text for input voice. A machine translation process is configured to output a corresponding second language text for an input first language text. A voice synthesis process is configured to output a corresponding voice for the input text. Each of the voice recognition and machine translation processes is configured to output identification information to identify a reading of a word when the text to be output contains the word having multiple readings. Each of the machine translation process and the voice synthesis process is configured to identify the reading of the word corresponding to the identification information based on the identification information when the input text includes the identification information.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a language processing system, a language processing method, and a language processing program. [Background technology]

[0002] With the rapid development of information and communication technology, communication support technologies such as speech recognition, machine translation, and speech synthesis have been widely put to practical use.

[0003] For example, as disclosed in Japanese Patent Laid-Open Publication No. 2007-073027 (Patent Document 1), a translation system that combines speech recognition, machine translation, and speech synthesis is known. This translation system performs translation based on the spoken content and meta-information of the spoken content. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2007-073027 A Summary of the Invention [Problem to be solved by the invention]

[0005] Generally, writing systems include ideographs (typically Chinese characters), where each character represents a meaning, and phonetic (or phonetic) characters, where each character represents a phoneme or syllable. Ideographs may have multiple pronunciations (readings).

[0006] The present invention aims to provide a technology for obtaining more accurate processing results for words that have multiple readings in a language processing system that combines at least two of speech recognition processing, machine translation processing, and speech synthesis processing. [Means for solving the problem]

[0007] According to an embodiment, there is provided a language processing system that combines at least two of a speech recognition process, a machine translation process, and a speech synthesis process. The language processing system is configured to transfer text between processes. The speech recognition process is configured to output a text corresponding to an input speech. The machine translation process is configured to output a text in a second language corresponding to an input text in a first language. The speech synthesis process is configured to output a speech corresponding to an input text. Each of the speech recognition process and the machine translation process is configured to output identification information for identifying the pronunciation of a word when the output text includes a word having multiple pronunciations. Each of the machine translation process and the speech synthesis process is configured to identify the pronunciation of a word corresponding to the identification information based on the identification information when the input text includes the identification information.

[0008] The voice recognition process and the machine translation process may each add identification information corresponding to a word that has multiple readings and output the result.

[0009] Each of the voice recognition process, the machine translation process, and the voice synthesis process may refer to a dictionary to identify the reading corresponding to the identification information.

[0010] Each of the voice recognition process and the machine translation process may refer to a dictionary to determine whether or not to output identification information.

[0011] The dictionary may be generated by extracting necessary information from a master dictionary that includes identification information, spelling in a first language, spelling in a second language, and pronunciation for a target word.

[0012] The identification information may be set to be unique among the words of interest.

[0013] The target words may include proper nouns.

[0014] The target word may include ideograms.

[0015] According to another embodiment, there is provided a language processing method in a language processing system that combines at least two of a speech recognition process, a machine translation process, and a speech synthesis process. The language processing system is configured to transfer text between processes. The speech recognition process of the language processing method includes a step of outputting a text corresponding to an input speech, and a step of outputting identification information for identifying the pronunciation of a word when the output text includes a word having multiple pronunciations. The machine translation process of the language processing method includes a step of outputting a text in a second language corresponding to an input text in a first language, a step of outputting identification information for identifying the pronunciation of a word when the output text includes a word having multiple pronunciations, and a step of identifying the pronunciation of a word corresponding to the identification information based on the identification information when the input text includes the identification information. The speech synthesis process of the language processing method includes a step of outputting a speech corresponding to an input text, and a step of identifying the pronunciation of a word corresponding to the identification information based on the identification information when the input text includes the identification information.

[0016] According to yet another embodiment, there is provided a program for causing a computer to execute the above-described language processing method. Effect of the Invention

[0017] According to the present invention, in a language processing system that combines at least two of speech recognition processing, machine translation processing, and speech synthesis processing, more accurate processing results can be obtained for words that have multiple readings. [Brief description of the drawings]

[0018] [Figure 1] 1 is a diagram for explaining an overview of a language processing method according to an embodiment of the present invention; [Diagram 2]11A and 11B are diagrams illustrating an application example of the language processing method according to the present embodiment. [Diagram 3] 1 is a diagram for explaining an example of a problem to be solved by the language processing method according to the present embodiment; [Figure 4] 11 is a diagram for illustrating an example of a method for generating a dictionary in the language processing method according to the present embodiment. FIG. [Diagram 5] 11A and 11B are diagrams illustrating an example of a process for outputting English speech in response to input of Japanese speech according to the present embodiment. [Figure 6] 11A and 11B are diagrams illustrating an example of a process for outputting Japanese speech in response to an input of English speech according to the present embodiment. [Figure 7] 1 is a schematic diagram showing an example of a hardware configuration of an information processing device constituting a language processing system according to an embodiment of the present invention; [Figure 8] 10 is a flowchart showing a processing procedure for voice recognition processing in the language processing method according to the present embodiment. [Figure 9] 10 is a flowchart showing a processing procedure for machine translation processing in the language processing method according to the present embodiment. [Figure 10] 10 is a flowchart showing a processing procedure for speech synthesis processing in the language processing method according to the present embodiment. [Figure 11] 13 is a diagram for illustrating another example of the method of generating a dictionary in the language processing method according to the present embodiment. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in detail with reference to the accompanying drawings, in which the same or corresponding parts in the drawings are designated by the same reference characters and will not be described repeatedly.

[0020] [A. Overview] Fig. 1 is a diagram for explaining an overview of a language processing method according to the present embodiment. Referring to Fig. 1, the language processing method according to the present embodiment is typically directed to a language processing system that combines at least two of a speech recognition process 10, a machine translation process 20, and a speech synthesis process 30.

[0021] The speech recognition process 10 outputs a corresponding text for an input speech. The machine translation process 20 outputs a corresponding text in a second language for an input text in a first language. The speech synthesis process 30 outputs a corresponding speech for an input text.

[0022] In the language processing system, text is passed between a speech recognition process 10, a machine translation process 20, and a speech synthesis process 30.

[0023] As an application example of the language processing method according to the present embodiment, the following configuration is assumed: Fig. 2 is a diagram showing an application example of the language processing method according to the present embodiment.

[0024] 2(A) shows an example of a configuration in which speech in a first language is input, translated into a second language, and then speech in the second language is output. In this example, speech in the first language is input to speech recognition processing 10, and text in the first language is output. The text in the first language is then input to machine translation processing 20, and text in the second language is output. Finally, the text in the second language is input to speech synthesis processing 30, and speech in the second language is output.

[0025] 2(B) shows an example of a configuration in which speech in a first language is input, translated into a second language, and text in the second language is output. In this example, speech in the first language is input to a speech recognition process 10, and text in the first language is output. The text in the first language is then input to a machine translation process 20, and text in the second language is output.

[0026] Fig. 2(C) shows an example of a configuration in which a speaker's voice is input and an arbitrary synthetic voice is output. For example, a case is assumed in which a voice with a voice quality different from that of the speaker is output. In this example, the speaker's voice is input to a voice recognition process 10, and text is output. The text is then input to a voice synthesis process 30, and an arbitrary synthetic voice is output.

[0027] 2(D) shows an example of a configuration in which a text in a first language is input, translated into a second language, and speech in the second language is output. In this example, the text in the first language is input to a machine translation process 20, and text in the second language is output. The text in the second language is then input to a speech synthesis process 30, and speech in the second language is output.

[0028] The technical concept of the present invention is not limited to the above-mentioned application examples.

[0029] [B. Assignment] Next, a new problem that the present inventors have discovered and that may arise in the above-mentioned language processing system will be described.

[0030] Fig. 3 is a diagram for explaining an example of a problem to be solved by the language processing method according to the present embodiment. Fig. 3(A) shows a problem in an example configuration in which Japanese speech is input and corresponding English speech is output. Fig. 3(B) shows a problem in an example configuration in which English speech is input and corresponding Japanese speech is output.

[0031] Figure 3 shows an example of processing text and speech including the word "Kobe Station" which can be read as "Kobeeki", "Goudoeki", "Kanbeeki", etc. However, the most frequently used reading of the word "Kobe Station" is "Kobeeki".

[0032] With reference to FIG. 3(A), first, assume that a speaker utters "Where is Kobe Station?". The speech recognition process 10 receives speech input from the speaker and outputs the text "Where is Kobe Station?". At this time, the pronunciation information of "Goudoeki" contained in the speech is discarded. Therefore, in the machine translation process 20, the frequently used "Kobe station" is selected as the translation corresponding to "Kobe Station". As a result, the machine translation process 20 outputs the text "Where is Kobe station?" as the translation corresponding to the input text. Finally, the speech synthesis process 30 outputs speech corresponding to the text output by the machine translation process 20.

[0033] In other words, when the user speaks "Where is Kobe station?", the incorrect speech output will be "Where is Kobe station?"

[0034] Similarly, referring to FIG. 3(B), assume that a speaker first utters "Where is Godo station?". The speech recognition processing 10 receives speech input from the speaker and outputs the text "Where is Godo station?". Next, the machine translation processing 20 outputs the text "Where is Kobe Station?" as a translation corresponding to the input text. At this time, the machine translation processing 20 only outputs text, and does not output pronunciation information. Therefore, in the speech synthesis processing 30, "KOBE-EKI", which is frequently used, is selected as the speech corresponding to "KOBE-EKI". As a result, the speech "Where is Kobe Station?" is output.

[0035] In other words, when you say "Where is Godo station?", the incorrect speech output will be "Where is Kobe station?"

[0036] [C. Solution] As described with reference to Fig. 3, since the text passed between processes does not have pronunciation information (reading), the input information may not be processed correctly. Therefore, the language processing method according to the present embodiment solves the above-mentioned problems by adopting a process of passing text with added pronunciation information between processes.

[0037] In the language processing method according to the present embodiment, in addition to the text, information for identifying pronunciation information (reading) can be added. The information for identifying pronunciation information (reading) may be in any format, but in the following description, an example including identification information called a "word ID" will be described. In other words, the identification information (word ID) is information for identifying the reading of a word that has multiple readings.

[0038] 1 again, the speech recognition process 10 outputs corresponding text and word ID for an input speech. That is, the speech recognition process 10 includes a process of outputting identification information (word ID) for identifying the pronunciation of a word when the output text includes a word with multiple pronunciations.

[0039] At this time, the voice recognition process 10 can determine the word ID to be output by referring to the voice recognition dictionary 60. That is, the voice recognition process 10 refers to the voice recognition dictionary 60 to identify the reading corresponding to the word ID.

[0040] Similarly, the machine translation process 20 outputs corresponding text (second language) and word ID for input text (first language). That is, the machine translation process 20 includes a process of outputting identification information (word ID) for identifying the pronunciation of a word when the output text includes a word with multiple pronunciations.

[0041] Furthermore, the machine translation process 20 can output corresponding text (second language) based on the word ID added to the input text (first language). That is, when the input text includes identification information (word ID), the machine translation process 20 includes a process of identifying the reading of the word corresponding to the identification information based on the identification information.

[0042] The machine translation process 20 can determine the word ID to be output, or can determine the processing content corresponding to the input word ID, by referring to the machine translation dictionary 70. In other words, the machine translation process 20 refers to the machine translation dictionary 70 to identify the reading corresponding to the word ID.

[0043] The speech synthesis process 30 outputs speech based on the added word ID for the input text. That is, when the input text includes identification information (word ID), the speech synthesis process 30 includes a process of identifying the pronunciation of the word corresponding to the identification information based on the identification information.

[0044] The voice synthesis process 30 can determine the processing content corresponding to the input word ID by referring to the voice synthesis dictionary 80. That is, the voice synthesis process 30 refers to the voice synthesis dictionary 80 to identify the reading corresponding to the word ID.

[0045] In any of the processes, the word ID may not be always output, but may be output only when necessary.

[0046] [D. Dictionary Generation] The dictionary for speech recognition 60, the dictionary for machine translation 70, and the dictionary for speech synthesis 80 may be generated by extracting necessary information from the master dictionary 50. An example of a method for generating the dictionary for speech recognition 60, the dictionary for machine translation 70, and the dictionary for speech synthesis 80 using the master dictionary 50 will be described below.

[0047] Fig. 4 is a diagram for explaining an example of a method for generating a dictionary in the language processing method according to the present embodiment. Fig. 4 shows, as an example, a master dictionary 50 assuming translation between Japanese and English. More specifically, referring to Fig. 4, master dictionary 50 includes a word ID column 51, a Japanese notation column 52, a Japanese reading column 53, and an English notation column 54.

[0048] Identification information for identifying the reading of a target word (in this example, "Kobe Station") is stored in the word ID column 51. In the example shown in Fig. 4, "001" is assigned to the reading "KOBE-EKI", "002" is assigned to the reading "GOUDO-EKI", and "003" is assigned to the reading "KANBAE-KI".

[0049] The identification information stored in the word ID column 51 may be set to be unique among multiple readings of the corresponding word. In other words, the identification information (word ID) may be set to be unique among the target words. Therefore, for example, the word ID "001" may be reused for other words.

[0050] However, the identification information stored in the word ID string 51 may be set to be unique within the scope of the master dictionary 50.

[0051] Character strings expressed using kanji characters etc. are stored in the Japanese notation column 52. That is, the Japanese notation column 52 stores notations of target words in the first language.

[0052] The Japanese pronunciation column 53 stores the pronunciation of the corresponding word in Japanese. That is, the Japanese pronunciation column 53 stores the pronunciation of the target word.

[0053] An English character string corresponding to a corresponding word is stored in the English notation column 54. That is, the English notation column 54 stores notations in the second language for the target word.

[0054] Necessary information is extracted from the information contained in the master dictionary 50, and a dictionary for speech recognition 60, a dictionary for machine translation 70, and a dictionary for speech synthesis 80 are generated.

[0055] More specifically, the Japanese speech recognition dictionary 60J and the Japanese speech synthesis dictionary 80J include a word ID column 51, a Japanese spelling column 52, and a Japanese reading column 53 of the master dictionary 50.

[0056] The machine translation dictionary 70 includes a word ID column 51 of the master dictionary 50 , a Japanese notation column 52 , and an English notation column 54 .

[0057] The English speech recognition dictionary 60E and the English speech synthesis dictionary 80E include the word ID column 51 and the English spelling column 54 of the master dictionary 50.

[0058] Although FIG. 4 shows the contents of the master dictionary 50 for the word "Kobe Station," other words can also be included in the master dictionary 50.

[0059] Also, while FIG. 4 shows an example of a master dictionary 50 that takes into account machine translation between Japanese and English, any language pair may be used, and a master dictionary 50 may be configured that takes into account machine translation between three or more languages.

[0060] Furthermore, although an example is shown in which the speech recognition dictionary 60, the machine translation dictionary 70, and the speech synthesis dictionary 80 are generated independently, the master dictionary 50 itself may be shared.

[0061] As described above, words with multiple readings are registered in the master dictionary 50 in association with each reading. Such target words are generally ideograms. Target words typically include proper nouns such as "Kobe Station."

[0062] [E. Processing example] Next, an example of processing using the speech recognition dictionary 60, the machine translation dictionary 70, and the speech synthesis dictionary 80 will be described.

[0063] FIG. 5 is a diagram showing an example of processing in which English speech is output in response to input of Japanese speech according to the present embodiment. With reference to FIG. 5, it is assumed that a speaker first utters "Where is Kobe Station?". The speech recognition processing 10 receives speech input from the speaker and outputs the text "Where is Kobe Station?". At this time, the speech recognition dictionary 60J is referenced and it is found that "Kobe Station" has multiple readings, so a word ID ("002" in this example) corresponding to "Goudoeki" included in the input speech is added to the output text. That is, the speech recognition processing 10 adds identification information (word ID) corresponding to a word having multiple readings and outputs the text.

[0064] Next, when "Kobe Station" and a word ID ("002" in this example) are input, the machine translation process 20 refers to the machine translation dictionary 70 and selects "Godo station", which is a translation of the reading corresponding to the word ID, from among the multiple readings of "Kobe Station". The machine translation process 20 outputs text including the selected "Godo station". At this time, the corresponding word ID may be added to the output text.

[0065] The speech synthesis process 30 outputs a speech corresponding to the text output by the machine translation process 20. At this time, the speech synthesis dictionary 80E may be referenced to look up or determine a speech corresponding to the added word ID.

[0066] Through this series of processes, the correct speech output, "Where is Godo station?" can be achieved in response to the speech request, "Where is Godo station?"

[0067] Fig. 6 is a diagram showing a processing example in which Japanese voice is output in response to an English voice input according to the present embodiment. With reference to Fig. 6, it is assumed that a speaker first utters "Where is Godo station?". The voice recognition processing 10 receives voice input from the speaker and outputs the text "Where is Godo station?". At this time, the voice recognition dictionary 60E may be referenced and a word ID corresponding to "Godo station" may be added to the output text.

[0068] Next, when "Godo station" and a word ID ("002" in this example) are input, the machine translation process 20 refers to the machine translation dictionary 70 and finds that one of the translations corresponding to the multiple readings of "Kobe Station" is "Godo Station." The machine translation process 20 then outputs text including "Kobe Station," which is the translation of "Godo Station," and adds a word ID ("002" in this example) for identifying "Godo Station" to the text. That is, the machine translation process 20 adds and outputs identification information (word ID) corresponding to a word with multiple readings.

[0069] The speech synthesis process 30 outputs a speech corresponding to the text output by the machine translation process 20. At this time, the speech synthesis dictionary 80J is referenced to select "GOUDO-EKI", which is the reading corresponding to the word ID added to the input text.

[0070] Through this series of processes, the system can output the correct speech, "Where is Godo station?" in response to the question, "Where is Godo station?"

[0071] In addition, since the voice synthesis process 30 shown in FIG. 5 and the voice recognition process 10 shown in FIG. 6 process text (English) having pronunciation information (reading), the processes of referencing the voice synthesis dictionary 80E and the voice recognition dictionary 60E, respectively, may be omitted.

[0072] Although an example in which a word ID is added to a text has been described, it is also possible to output only the word ID instead of the text indicating the target word. In this case, the value of the word ID is set so that it is unique among all words in the master dictionary 50.

[0073] Also, word IDs may be set only for proper nouns (personal names, place names, names, etc. that have multiple readings) and words that need to be managed by the entire language processing system. In this case, the processing mode may change, but details will be described later.

[0074] [F. Language Processing Systems] Next, an example of a hardware configuration of the language processing system according to the present embodiment will be described.

[0075] 7 is a schematic diagram showing an example of a hardware configuration of an information processing device 100 constituting a language processing system according to the present embodiment. Typically, information processing device 100 can be realized using a general-purpose computer.

[0076] 7, the information processing device 100 includes, as main hardware components, a central processing unit (CPU) 102, a graphics processing unit (GPU) 104, a main memory 106, a display 108, a network interface (I / F) 110, a secondary storage device 112, an input device 122, and an optical drive 124. These components are connected to each other via an internal bus 128.

[0077] The CPU 102 and / or the GPU 104 are processors that execute the language processing method according to the present embodiment. A plurality of CPUs 102 and GPUs 104 may be provided, and each may have a plurality of cores.

[0078] The main memory 106 is a storage area that temporarily stores (or caches) program code, work data, etc. when the processor (CPU 102 and / or GPU 104) executes processing, and is composed of volatile memory devices such as DRAM (dynamic random access memory) and SRAM (static random access memory).

[0079] The display 108 is a display unit that outputs a user interface relating to processing, processing results, and the like, and is configured, for example, with an LCD (liquid crystal display) or an organic EL (electroluminescence) display.

[0080] The network interface 110 exchanges data with any information processing device on the Internet or an intranet. As the network interface 110, any communication method such as Ethernet (registered trademark), wireless LAN (local area network), Bluetooth (registered trademark), etc. can be adopted.

[0081] The input device 122 is a device that accepts instructions and operations from a user, and is composed of, for example, a keyboard, a mouse, a touch panel, a pen, etc. The input device 122 may also include a sound collection device for collecting audio signals, and may also include an interface for accepting input of audio signals collected by the sound collection device.

[0082] The optical drive 124 reads out information stored on an optical disk 126, such as a compact disc read only memory (CD-ROM) or a digital versatile disc (DVD), and outputs the information to other components via an internal bus 128. The optical disk 126 is an example of a non-transitory recording medium, and is distributed in a state in which an arbitrary program is stored in a non-volatile manner. The optical drive 124 reads out a program from the optical disk 126 and installs it in the secondary storage device 112 or the like, so that the computer functions as the information processing device 100. Therefore, the subject of the present invention may be the program itself installed in the secondary storage device 112 or the like, or a recording medium such as the optical disk 126 that stores a program for realizing the functions and processing according to the present embodiment.

[0083] FIG. 7 shows an optical recording medium such as an optical disk 126 as an example of a non-transient recording medium, but the present invention is not limited to this. Alternatively, a semiconductor recording medium such as a flash memory, a magnetic recording medium such as a hard disk or a storage tape, or a magneto-optical recording medium such as an MO (magneto-optical disk) may be used.

[0084] The secondary storage device 112 stores programs and data necessary for the computer to function as the information processing device 100. For example, the secondary storage device 112 is configured with a non-volatile storage device such as a hard disk or a solid state drive (SSD).

[0085] More specifically, the secondary storage device 112 stores, in addition to an OS (operating system) not shown, a speech recognition program 114 for implementing speech recognition processing 10, a machine translation program 116 for implementing machine translation processing 20, a speech synthesis program 118 for implementing speech synthesis processing 30, a speech recognition dictionary 60, a machine translation dictionary 70, and a speech synthesis dictionary 80.

[0086] The speech recognition program 114, the machine translation program 116, and the speech synthesis program 118 correspond to programs for causing the information processing device 100 to execute the language processing method according to the present embodiment. The speech recognition program 114, the machine translation program 116, and the speech synthesis program 118 can also be realized using machine learning. In this case, these programs include a trained model.

[0087] Some of the libraries and functional modules required when the processor (CPU 102 and / or GPU 104) executes a program may be replaced with libraries or functional modules provided as standard by the OS. In this case, the program alone does not include all of the program modules required to realize the corresponding function, but the desired processing can be realized by installing it under the execution environment of the OS. Even a program that does not include some of such libraries or functional modules may be included in the technical scope of the present invention.

[0088] These programs may not only be distributed by being stored in any of the above-mentioned recording media, but may also be distributed by being downloaded from a server device or the like via the Internet or an intranet.

[0089] FIG. 7 shows an example of configuring information processing device 100 using a single computer, but the present invention is not limited to this. A plurality of computers connected via a computer network may cooperate explicitly or implicitly to realize the language processing method according to this embodiment.

[0090] All or part of the functions realized by the processor (CPU 102 and / or GPU 104) executing the programs may be realized using a hard-wired circuit such as an integrated circuit. For example, they may be realized using an application specific integrated circuit (ASIC) or a field-programmable gate array (FPGA).

[0091] A person skilled in the art would be able to realize the information processing device 100 according to the present embodiment by appropriately using technology appropriate for the era in which the present invention is implemented.

[0092] For convenience of explanation, an example will be shown in which the speech recognition process 10, the machine translation process 20 and the speech synthesis process 30 are executed using the same information processing device 100, but each process may be realized using different hardware.

[0093] [G. Processing Procedure] Next, the processing procedure of the language processing method according to the present embodiment will be described. The language processing method according to the present embodiment combines some or all of the speech recognition process 10, the machine translation process 20, and the speech synthesis process 30, but each process can be executed independently, so the following description will focus on each process.

[0094] 8 is a flowchart showing the processing procedure of the speech recognition process 10 of the language processing method according to the present embodiment. Each step shown in FIG. 8 is typically realized by the processor (CPU 102 and / or GPU 104) of the information processing device 100 executing the speech recognition program 114.

[0095] 8, information processing device 100 determines whether or not any voice has been input (step S100). If no voice has been input (NO in step S100), the process of step S100 is repeated.

[0096] If any voice has been input (YES in step S100), the information processing device 100 generates text corresponding to the input voice (step S102). In this way, the voice recognition process 10 includes a process of outputting text corresponding to the input voice.

[0097] Then, the information processing device 100 refers to the voice recognition dictionary 60 to determine whether or not a word with multiple readings exists in the text corresponding to the input voice (step S104). That is, the information processing device 100 refers to the voice recognition dictionary 60 to determine whether or not to output identification information (word ID).

[0098] If the text contains a word with multiple readings (YES in step S104), the information processing device 100 refers to the speech recognition dictionary 60 to determine a word ID indicating the reading corresponding to the input speech (step S106), and adds the word ID to the text corresponding to the input speech (step S108). Thus, the speech recognition process 10 includes a process of outputting identification information (word ID) for identifying the reading of a word when the output text contains a word with multiple readings.

[0099] If there are no words with multiple readings in the text (NO in step S104), the processes in steps S106 and S108 are skipped.

[0100] Then, the information processing device 100 outputs the processing result (the text, or the text to which the word ID has been added) (step S110), and the processing ends.

[0101] 9 is a flowchart showing the processing procedure of machine translation process 20 of the language processing method according to the present embodiment. Each step shown in FIG. 9 is typically realized by the processor (CPU 102 and / or GPU 104) of information processing device 100 executing machine translation program 116.

[0102] 9, information processing device 100 determines whether or not any text has been input (step S200). If no text has been input (NO in step S200), the process of step S200 is repeated.

[0103] If any text has been input (YES in step S200), information processing device 100 determines whether or not a word ID has been added to the input text (step S202).

[0104] If a word ID has been added to the input text (YES in step S202), the information processing device 100 refers to the machine translation dictionary 70 to determine the reading corresponding to the word ID (step S204). Then, the information processing device 100 generates text (translation) in another language corresponding to the input text, reflecting the determined reading (step S206). In this way, the machine translation process 20 includes a process of outputting text in a second language corresponding to the input text in the first language, and also includes a process of specifying the reading of the word corresponding to the identification information based on the identification information when the input text in the first language includes identification information.

[0105] Then, the information processing device 100 adds the input word ID to the generated text in another language (step S208).Then, the information processing device 100 outputs the processing result (text in another language, or text in another language to which the word ID has been added) (step S210), and the processing ends.

[0106] On the other hand, if a word ID is not added to the input text (NO in step S202), the information processing device 100 generates a text (translation) in another language corresponding to the input text (step S212). In this way, the machine translation process 20 includes a process of outputting a text in a second language corresponding to the input text in a first language.

[0107] Then, the information processing device 100 refers to the machine translation dictionary 70 to determine whether or not the generated text (translation) in another language contains words with multiple readings (step S214). That is, the information processing device 100 refers to the machine translation dictionary 70 to determine whether or not to output identification information (word ID).

[0108] If the text contains a word with multiple readings (YES in step S214), the information processing device 100 refers to the machine translation dictionary 70 to determine a word ID indicating the reading of the target word contained in the generated text (step S216), and adds the word ID to the generated text (step S218). Thus, the machine translation process 20 includes a process of outputting identification information (word ID) for specifying the reading of a word when the text to be output contains a word with multiple readings.

[0109] If there are no words with multiple readings in the text (NO in step S214), the processes in steps S216 and S218 are skipped.

[0110] Then, the information processing device 100 outputs the processing result (the text in the other language, or the text in the other language to which the word ID is added) (step S220), and the processing ends.

[0111] Fig. 10 is a flowchart showing the processing procedure of speech synthesis process 30 of the language processing method according to the present embodiment. Each step shown in Fig. 10 is typically realized by the processor (CPU 102 and / or GPU 104) of information processing device 100 executing speech synthesis program 118.

[0112] 10, information processing device 100 determines whether or not any text has been input (step S300). If no text has been input (NO in step S300), the process of step S300 is repeated.

[0113] If any text has been input (YES in step S300), information processing device 100 determines whether or not a word ID has been added to the input text (step S302).

[0114] If a word ID is added to the input text (YES in step S302), the information processing device 100 refers to the speech synthesis dictionary 80 and determines the pronunciation corresponding to the word ID (step S304). In this way, when the input text includes identification information (word ID), the speech synthesis process 30 includes a process of specifying the pronunciation of the word corresponding to the identification information based on the identification information.

[0115] Then, the information processing device 100 generates a voice corresponding to the input text, reflecting the determined reading (step S306).

[0116] On the other hand, if a word ID is not added to the input text (NO in step S302), the information processing device 100 generates a voice corresponding to the input text (step S308).

[0117] In this manner, the speech synthesis process 30 includes a process of outputting a voice corresponding to an input text.

[0118] Then, information processing device 100 outputs the generated voice (step S310), and the process ends.

[0119] [H. Variations in Processing Procedure] The process described in the above-mentioned Figures 8 to 10 is for a case where a word ID is set uniquely for a word that has multiple readings. On the other hand, as mentioned above, it is also possible to make all the added word IDs different.

[0120] Fig. 11 is a diagram for explaining another example of a method for generating a dictionary in the language processing method according to the present embodiment. With reference to Fig. 11, for example, in the master dictionary 50, a word ID is set uniquely for words that have multiple readings (in the example of Fig. 11, "Kobe Station" and "Shinjuku"). From such a master dictionary 50, a Japanese voice recognition dictionary 60J and a voice synthesis dictionary 80J, a machine translation dictionary 70, an English voice recognition dictionary 60E and a voice synthesis dictionary 80E are generated. In each dictionary that is generated, words that do not have multiple readings are also registered, but such words do not need to be assigned a word ID.

[0121] When such a word ID assignment form is adopted, the processes of speech recognition, machine translation, and speech synthesis will differ from those shown in Figures 8 to 10. Each process will be briefly described below with reference to Figures 5 and 6 again.

[0122] In speech recognition, an input speech signal is converted into text with reference to the speech recognition dictionary 60. Therefore, when the speech signal is converted into text, the word IDs are identified for words to which a word ID has been added. Therefore, the word IDs are added to the text that is the output of the speech recognition process 10. Note that words to which a word ID has been added in the speech recognition dictionary 60 (i.e., the master dictionary 50) are output together with the word ID, and words to which no word ID has been added are output without a word ID.

[0123] In the machine translation process 20, the input text is translated. At that time, a machine translation dictionary 70 is referenced based on the word ID added to the text, and a translation of the pronunciation specified by the word ID output from the speech recognition process 10 is selected. In a typical machine translation process 20, a morphological analysis is performed on the text (surface text) of the speech recognition result, and proper nouns are replaced with labels that are proper nouns, before the machine translation process is performed. Therefore, by replacing the results of the machine translation with the results obtained by searching the machine translation dictionary 70 using the word ID, it is possible to output a machine translation result that holds correct pronunciation information as a word ID.

[0124] The speech synthesis process 30 receives the text and word ID information output from the machine translation process 20 and refers to a speech synthesis dictionary 80, thereby enabling correct speech corresponding to the original speech input.

[0125] In this way, when word IDs are set uniquely in the master dictionary (i.e., in each corresponding dictionary), there is no need to search whether each word has multiple readings during machine translation and speech synthesis processes, making processing more efficient.

[0126] [I. Modifications] In the above-described embodiment, the speech recognition process 10, the machine translation process 20, and the speech synthesis process 30 respectively refer to the speech recognition dictionary 60, the machine translation dictionary 70, and the speech synthesis dictionary 80 to resolve the correspondence between the identification information (word ID) and the pronunciation, but this is not limiting and a machine-learned model including the contents of each dictionary may be used. In this case, the information of the speech recognition dictionary 60, the machine translation dictionary 70, and the speech synthesis dictionary 80 (or the master dictionary 50) will be included in the trained model.

[0127] In such a configuration, text and identification information (word ID), or identification information (word ID), are input into the trained model, and a result reflecting the reading corresponding to the identification information is output.

[0128] [J. Summary] According to the language processing method of this embodiment, the text transferred between processes includes identification information indicating pronunciation information (reading), making it possible to obtain more accurate processing results for words that have multiple readings.

[0129] The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is defined by the claims, not by the description of the embodiments described above, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0130] 10 speech recognition processing, 20 machine translation processing, 30 speech synthesis processing, 50 master dictionary, 51 word ID column, 52 Japanese notation column, 53 Japanese reading column, 54 English notation column, 60, 60E, 60J speech recognition dictionary, 70 machine translation dictionary, 80, 80E, 80J speech synthesis dictionary, 100 information processing device, 102 CPU, 104 GPU, 106 main memory, 108 display, 110 network interface, 112 secondary storage device, 114 speech recognition program, 116 machine translation program, 118 speech synthesis program, 122 input device, 124 optical drive, 126 optical disk, 128 internal bus.

Claims

1. A language processing system that combines at least two of a speech recognition process, a machine translation process, and a speech synthesis process, The language processing system is configured to transfer text between processes, The speech recognition process is configured to output a corresponding text for an input speech, The machine translation process is configured to output a corresponding text in a second language for an input text in a first language; The speech synthesis process is configured to output a corresponding speech for an input text, The unique identification information is shared within the language processing system based on a master dictionary in which the unique identification information and the pronunciation of the word are registered in association with each other, each of the speech recognition process and the machine translation process is configured to add, when an output text includes a word having a plurality of readings, identification information registered in the master dictionary for identifying the readings of the word to the output text; A language processing system, wherein each of the machine translation process and the speech synthesis process is configured to, when identification information is added to input text, determine the pronunciation of a word corresponding to the identification information based on registration information in the master dictionary.

2. The language processing system according to claim 1 , wherein each of the speech recognition process and the machine translation process adds identification information corresponding to a word having multiple readings.

3. The language processing system includes the master dictionary, 3. The language processing system according to claim 1, wherein each of the speech recognition process, the machine translation process, and the speech synthesis process identifies the reading corresponding to the identification information by referring to a dictionary for each process generated by extracting necessary information from the master dictionary.

4. 4. The language processing system according to claim 3, wherein each of the speech recognition process and the machine translation process refers to a dictionary for each process to determine whether or not the identification information should be output.

5. A language processing method in a language processing system that combines at least two of a speech recognition process, a machine translation process, and a speech synthesis process, comprising the steps of: The language processing system is configured to transfer text between processes, The unique identification information is shared within the language processing system based on a master dictionary in which the unique identification information and the pronunciation of the word are registered in association with each other, The speech recognition process of the language processing method includes: outputting a text corresponding to the input speech; and when the output text includes a word having multiple readings, outputting identification information registered in the master dictionary for identifying the readings of the word; The machine translation process of the language processing method includes: outputting a corresponding text in a second language for the input text in the first language; outputting, when the output text includes a word having multiple readings, identification information for identifying the reading of the word; and when the input text includes identification information, determining the reading of the word corresponding to the identification information based on the registration information of the master dictionary; The speech synthesis process of the language processing method includes: outputting a corresponding voice for the input text; When the input text contains identification information, the language processing method further comprises a step of identifying the pronunciation of a word corresponding to the identification information based on registration information in the master dictionary.

6. A program for causing a computer to execute the language processing method according to claim 5.

Images