JP2004252121A - Language processing apparatus and language processing method, and program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To make a registered word commonly usable when the word for a plurality of applications is registered. <P>SOLUTION: When voices are inputted to a microphone 51, an analog-to-digital conversion section 52 converts a voice signal which is an analog signal into voice data which is a digital signal. A characteristic amount extraction section 53 extracts a characteristic amount from voice data and supplies the same to a matching section. The matching section 54 references the task corresponding to the application, recognizes the voice based on the characteristic amount and registers the word in a common dictionary section 55. Also, the matching section 54 changes the word registered in the section 55. The matching section 54 reflects the content of the section 55 to the respective tasks. The invention is applicable to a robot control system. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a language processing apparatus, a language processing method, a program, and a recording medium, and more particularly to, for example, a language processing apparatus, a language processing method, and a program that allow a registered word to be commonly recognized by a plurality of applications.
[0002]
[Prior art]
Speech recognition includes isolated word recognition for recognizing a single word and continuous word recognition for recognizing a word string composed of a plurality of words. In conventional continuous word recognition, having a "language model of a database of the ease of connection between words" prevents the generation of "similar but messy word strings" as recognition results. In.
[0003]
However, since only information on words that can be recognized from the beginning (hereinafter, appropriately referred to as known words) is described in the language model, words registered later (hereinafter, appropriately referred to as registered words) are correctly recognized. It was difficult to do. In other words, in isolated word recognition, if a word is registered in the recognition dictionary, that word will be recognized in the future. However, in continuous word recognition, registration in the dictionary alone is not sufficient. Although it is necessary to reflect it, it was generally difficult to reflect it on the language model.
[0004]
Therefore, it has been proposed to classify the registered words into categories such as "person name" and "place name", prepare a recognition grammar corresponding to the category, and recognize speech (for example, see Patent Document 1).
[0005]
Further, in a system in which a plurality of applications that use voice recognition and a variable number of applications exist, when a word registered in one application is reflected in another application, a problem different from the case of a single application occurs. . For example, if word registration is performed only for an application that has already been started, it is difficult to reflect the registered word on an application that has been started or installed after registration, unlike the case where there is only one application. There were challenges.
[0006]
Further, when there are a plurality of applications, it is troublesome to delete the same registered word many times in a plurality of applications. Further, when there are a plurality of applications, it is easy to delete all the registered words, but it is difficult to delete only a part of the words or change the pronunciation.
[0007]
That is, when there is one application, for example, the registered word to be deleted or changed can be specified by information such as “the word registered in the nth time” or “the nth entry in the recognition dictionary”, but there are a plurality of applications. In such a case, it is difficult to specify the “word registered in the nth time” or “the number of the dictionary entry added” depending on each application.
[0008]
When there are a plurality of applications, the registered word can be specified by pronunciation. However, when the registered word is specified by pronunciation, the homonym may be deleted or changed.
[0009]
Therefore, it has been proposed that instead of each application performing voice recognition individually, a module called “voice commander” performs voice recognition for all applications and transfers the recognition result to each application (for example, see Patent Reference 1).
[0010]
[Patent Document 1]
JP 2001-216128 A
[0011]
[Problems to be solved by the invention]
However, in the invention described in Patent Literature 1, it is necessary for the “voice commander” to possess a recognition dictionary and a language model corresponding to each application. In other words, when developing a "voice commander", it is necessary to suppose what applications will be used at the same time, and to prepare a recognition dictionary and language model suitable for those applications. On the other hand, there is a problem that it is difficult to reflect registered words.
[0012]
The present invention has been made in view of such a situation, and aims to allow a registered word to be commonly used by a plurality of applications.
[0013]
[Means for Solving the Problems]
The language processing apparatus of the present invention includes a registered dictionary storage unit that stores a registered dictionary in which words are registered, and a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered. Constructing means for constructing based on the registered dictionary.
[0014]
A processing unit for adding, deleting, or changing words to the registered dictionary, and a deleting unit for deleting words in the dedicated dictionary are further provided. After all words registered in the dedicated dictionary are deleted, The construction means may reconstruct the dedicated dictionary based on the registered dictionary in which the word has been added, deleted, or changed.
[0015]
The dedicated dictionary includes at least a fixed dictionary in which predetermined words are registered in advance and a variable dictionary in which registered words are variable, and the construction unit is configured to construct a variable dictionary among the dedicated dictionaries. Can be.
[0016]
The dedicated dictionary further includes a category table in which categories of words are registered, and the constructing unit registers words of the categories registered in the category table among words of the registered dictionaries in the variable dictionary. Can be built.
[0017]
There are a plurality of applications, and the construction means can construct a dedicated dictionary for each of the plurality of applications.
[0018]
The language processing method of the present invention includes a registered dictionary storage step for storing a registered dictionary in which words are registered, and a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered. And a construction step of constructing based on the registered dictionary.
[0019]
The program recorded on the recording medium of the present invention constructs a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered based on the registered dictionary in which the words are registered. It is characterized by including a construction step.
[0020]
A program of the present invention causes a computer to execute a construction step of constructing a dedicated dictionary dedicated to an application in which words to be subjected to language processing used in an application are registered based on the registered dictionary in which the words are registered. It is characterized by the following.
[0021]
In the present invention, a registered dictionary in which words are registered is stored, and a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered is constructed based on the registered dictionary.
[0022]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 shows a configuration example of a robot control system 1 to which the present invention is applied.
[0023]
In the robot control system 1, the voice recognition engine unit 11 recognizes the input voice data, and generates a word string corresponding to the voice data as a recognition result. The speech recognition engine unit 11 sends the recognition result to the name registration application unit 21. ₁ , Chat application unit 21 ₂ , Voice commander application section 21 ₃ ,..., Other application units 21 _M And to the application management unit 31.
[0024]
Name registration application section 21 ₁ , Chat application unit 21 ₂ , Voice commander application section 21 ₃ ,..., Other application units 21 _M Performs various processes based on the recognition result supplied from the speech recognition engine unit 11.
[0025]
Name registration application section 21 ₁ Registers the robot name, the user name, and the like by voice based on the recognition result supplied from the voice recognition engine unit 11, and the other application units include the name registration application unit 21 ₁ Controls the operation of the robot in response to the user's utterance using the name registered by the user.
[0026]
Therefore, the chat application unit 21 ₂ , Voice commander application section 21 ₃ ,... And other application units 21 _M Is performed by the name registration application unit 21. ₁ It is necessary to correspond to the robot name, user name, etc. registered in.
[0027]
Chat application unit 21 ₂ Makes the robot chat with the user by voice, and the voice commander application unit 21 ₃ Causes the robot to perform an operation corresponding to the utterance from the user. For example, the voice commander application unit 21 ₃ Moves the robot forward in response to an utterance from the user such as "SDR (robot name), move forward!"
[0028]
Note that an arbitrary number of application units can be prepared. Hereinafter, the name registration application unit 21 ₁ , Chat application unit 21 ₂ , Voice commander application section 21 ₃ ,... And other application units 21 _M Are not collectively referred to as the application unit 21 as appropriate.
[0029]
The application management unit 31 instructs the application unit 21 to start and end based on the recognition result supplied from the speech recognition engine unit 11. For example, when the recognition result of “activate voice commander” is supplied from the voice recognition engine unit 11, the application management unit 31 ₃ Start At this time, a plurality of application units may be started simultaneously.
[0030]
Further, the application unit 21 and the application management unit 31 issue a task switching command to the speech recognition engine unit 11, and a corresponding task (to be described later with reference to FIG. 2) is stored in the speech recognition engine unit 11. Is controlled to be valid (active) or invalid (deactive).
[0031]
FIG. 2 shows the configuration of the speech recognition engine unit 11. The utterance of the user is input to the microphone 51, where the utterance is converted into an audio signal as an electric signal. The microphone 51 supplies this audio signal to an AD (Analog Digital) converter 52. The AD converter 52 samples and quantizes the audio signal that is an analog signal from the microphone 51 and converts it into audio data that is a digital signal. This audio data is supplied to the feature amount extraction unit 43.
[0032]
The feature amount extraction unit 53 extracts feature parameters such as a spectrum, a power linear prediction count, a cepstrum count, and a line spectrum pair from the audio data from the AD conversion unit 52 for each appropriate frame, and supplies the extracted feature parameters to the matching unit 54. I do.
[0033]
The matching unit 54 performs a phonemic typewriter task 71 based on the feature parameters from the feature amount extraction unit 53. ₁ , Application switching task 71 ₂ , Name registration task 71 ₃ , Chat task 71 ₄ , Voice commander task 71 ₅ , ..., and other tasks 71 _N Of these, for each task enabled at that time, a word string closest to the voice (input voice) input to the microphone 51 is obtained as a recognition result while referring to a database inside the task as necessary. . The matching unit 54 supplies the recognition result to the application unit 21 and the application management unit 31 corresponding to each task.
[0034]
Note that a task is a set of data necessary for performing speech recognition. That is, it is a program for accessing the data part when the speech recognition engine unit 11 is classified into a program part for performing matching or the like, and a data part such as an acoustic model, a language model, a recognition dictionary, and the like. .
[0035]
Therefore, even when a plurality of applications perform speech recognition using different acoustic models, language models, and dictionaries, a single speech recognition engine unit can be provided by preparing a plurality of tasks. Details of the inside of the task will be described later with reference to FIG.
[0036]
Phonemic Typewriter Task 71 ₁ Is a task that works as a phoneme typewriter, and is enabled by a command from the speech recognition engine unit 11. With this phoneme typewriter, the matching unit 54 obtains a phoneme sequence and also obtains pronunciation in kana notation for any input speech. For example, from the sound "Your name is SDR", "k / i / m / i / n / o / n / a / m / a / e / w / a / e / s / u / d / I: / a: / r / u / d / a / y / o ”(“ i: ”and“ a: ”are the long sounds of“ i ”and“ a ”, respectively), and“ Kimino Namae ” We get Kana notation called "WSDD Dayo". The phoneme sequence and the kana notation are used in the unknown word acquisition unit 56.
[0037]
Application switching task 71 ₂ Is a task corresponding to the application management unit 31, and is activated when a task switching command is supplied from the application management unit 31 after the application management unit 31 is activated. Application switching task 71 ₂ Accordingly, the matching unit 54 recognizes a voice corresponding to an activation of an application unit such as “launch a chat application”, “launch a voice commander”, “launch a name registration”, or an end command.
[0038]
Name registration task 71 ₃ Is the name registration application unit 21 ₁ This is a task corresponding to the name registration application unit 21 according to a command from the application management unit 31. ₁ Is activated, the name registration application section 21 ₁ When a task switching command is supplied from, it is enabled. Name registration task 71 ₃ For example, the matching unit 54 outputs a voice corresponding to a name such as "Your name is <unknown word representing a robot name>." And "My name is <unknown word representing a person name." Recognize.
[0039]
Chat task 71 ₄ , Voice commander task 71 ₅ , ..., and other tasks 71 _N Is the chat application unit 21 ₂ , Voice commander application section 21 ₃ ,..., Other application units 21 _M The task is enabled when a task switching command is supplied from the corresponding application unit after the corresponding application unit is activated by a command from the application management unit 31.
[0040]
The matching unit 54 includes a chat task 71. ₄ Thereby, for example, it is possible to recognize an utterance as a chat from the user, "SDR (robot name), when did it happen?" The matching unit 54 includes a voice commander task 71. ₅ Thereby, for example, it is possible to recognize an utterance as a command from the user, "SDR (robot name), move forward one step".
[0041]
Further, the matching unit 54 reflects words registered in a common dictionary unit 55 described later in each task.
[0042]
In the following, a phonemic typewriter task 71 will be described. ₁ , Application switching task 71 ₂ , Name registration task 71 ₃ , Chat task 71 ₄ , Voice commander task 71 ₅ , ..., and other tasks 71 _N Are not collectively referred to as a task 71 when it is not necessary to individually distinguish them.
[0043]
The common dictionary unit 55 stores a common dictionary as a dictionary of words commonly used in the task 71. In the common dictionary stored in the common dictionary unit 55, pronunciation information and category information are described for all words registered therein. For example, when the proper noun “SDR (robot name)” is registered in the common dictionary, the pronunciation (phonetic information) of “SDR” and the category “_robot name_” are described in the common dictionary. Details will be described later with reference to FIG.
[0044]
The unknown word acquisition unit 56 performs a phoneme typewriter task 71 on a word (unknown word) such as a name that is not registered in a recognition dictionary (fixed word dictionary 131 described later in FIG. 6). ₁ The phoneme sequence and the kana notation supplied from the matching unit 54 are stored, and thereafter, the voice of the word can be recognized (can be distinguished from other voices).
[0045]
That is, the unknown word acquisition unit 56 performs the task 71 for the phoneme typewriter. ₁ Is classified into several clusters. Each cluster has an ID, a representative phoneme sequence, and a representative kana notation, and is managed by the ID.
[0046]
FIG. 3 shows the state of the cluster of the unknown word acquisition unit 56.
[0047]
When three voices “Aka”, “Ao”, and “Midori” are input, the unknown word acquisition unit 56 converts the three input voices into “Aka” cluster 91 and “Ao” cluster corresponding to the three voices, respectively. 92 and a “green” cluster 93, and each cluster has a representative phoneme sequence (“a / k / a”, “a / o”, “m / i / d / o / r / i ″), representative kana notation (“red”, “blue”, “midori” in the example of FIG. 3), and ID (“1” in the example of FIG. 3). , "2", "3").
[0048]
Here, when the voice “red” is input again, the corresponding cluster already exists, so the unknown word acquisition unit 56 classifies the input voice into the “red” cluster 91 and does not generate a new cluster. On the other hand, when the voice of “kuro” is input, there is no corresponding cluster, so the unknown word acquisition unit 56 newly generates a “kuro” cluster 94 corresponding to “kuro”, and 3, a representative phoneme sequence (“k / u / r / o” in the example of FIG. 3), a representative kana notation (“black” in the example of FIG. 3), and an ID (FIG. In the case of the example, “4”) is added.
[0049]
By using this method, the accuracy of the representative phoneme sequence and the representative kana pronunciation of each cluster can be improved by the user inputting the same voice many times. For example, it is assumed that, when "Midori" is input once, the phonological typewriter misrecognizes and outputs a phonological sequence "m / e / r / a / a" and a kana pronunciation "Meraa". . After that, by repeating the utterance "Midori" many times, the phonemic sequence and Kana pronunciation may converge to the correct values ("m / i / d / o / r / i" and "Midori"). There is. Details of such a word acquisition process are disclosed in Japanese Patent Application Nos. 2001-097842 and 2001-382579 previously proposed by the present applicant.
[0050]
Next, a robot control process in the robot control system 1 of FIG. 1 will be described with reference to FIGS. This process is started when the robot control system 1 is started by a user.
[0051]
In step S1, the speech recognition engine unit 11 starts up, and proceeds to step S2. In step S2, the voice recognition engine unit 11 stores the contents of the common dictionary unit 55 (common dictionary) stored in a storage unit (not shown) at the end of the previous robot control system 1 (processing in step S17 described later). And the state of the cluster of the unknown word acquisition unit 56 are loaded. When the state of the common dictionary and the cluster is not stored in the storage unit, the common dictionary unit 55 and the unknown word acquiring unit 56 leave no cluster entry. In the case where the state of the cluster is stored in the storage unit but the state of the common dictionary is not stored, only the common dictionary is initialized (the state where there is no entry). Conversely, when the state of the common dictionary is stored but the state of the cluster is not stored, the entry derived from the cluster (the entry in which the cluster ID is described in FIG. 24) is deleted from the common dictionary, and the kana pronunciation is performed. The entry of origin (the entry in which kana pronunciation is described in FIG. 24) is left.
[0052]
After the processing in step S2, the process proceeds to step S3, where the speech recognition engine unit 11 executes the phoneme typewriter task 71. ₁ And enable phonetic typewriter task 71 ₁ Is made available for speech recognition, and the process proceeds to step S4. In step S4, the application management unit 31 is activated, and proceeds to step S5.
[0053]
In step S5, the application management unit 31 determines that the corresponding task is the application switching task 71. ₂ Is enabled, and the process proceeds to step S6. In step S <b> 6, the voice recognition engine unit 11 recognizes a start command of the application unit 21 input by voice to the microphone 51 and supplies a recognition result to the application management unit 31. The details of the voice recognition processing will be described later with reference to the flowchart in FIG.
[0054]
After the process in step S6, the process proceeds to step S7 in FIG. 5, and the application management unit 31 determines the name registration application unit 21 from the recognition result supplied from the speech recognition engine unit 11. ₁ It is determined whether or not to start up, and the name registration application unit 21 ₁ Is started (for example, when the recognition result is "start name registration"), the process proceeds to step S8.
[0055]
In step S8, the application management unit 31 sends the name registration application unit 21 ₁ Start. After the processing in step S8, the process proceeds to step S9, where the name registration application unit 21 ₁ Performs a name registration process. The details of the name registration process will be described later with reference to the flowchart of FIG.
[0056]
In step S7, the application management unit 31 sends the name registration application unit 21 ₁ If not, the process proceeds to step S10, and based on the recognition result by the speech recognition engine unit 11, the chat application unit 21 ₂ Is determined. In step S10, the application management unit 31 ₂ Is started (for example, when the recognition result is “start chat”), the process proceeds to step S11, and the chat application unit 21 is started. ₂ Start.
[0057]
After the processing in step S11, the process proceeds to step S12, where the chat application unit 21 ₂ Performs chat processing. Details of the chat processing will be described later with reference to a flowchart of FIG.
[0058]
In step S10, the application management unit 31 ₂ If it is determined not to activate the voice commander application unit 21 based on the recognition result by the voice recognition engine unit 11, ₃ Is determined. In the processing of step S13, the application management unit 31 ₃ Is started (for example, when the recognition result is “voice commander activation”), the process proceeds to step S14, and the voice commander application unit 21 is started. ₃ Start.
[0059]
After the processing of step S14, the process proceeds to step S15, where the voice commander application unit 21 ₃ Performs voice commander processing. Details of the voice commander processing will be described later with reference to the flowchart of FIG.
[0060]
In step S13, the application management unit 31 sends the voice commander application unit 21 ₃ Is determined not to be activated, the recognition result by the voice recognition engine unit 11 is incorrect (there may be an utterance other than the application switching), so the process returns to step S6 in FIG. 4, and the voice recognition engine unit 11 A process for recognizing the input voice is performed.
[0061]
Thus, the application management unit 31 activates the application unit 21 according to the recognition result by the speech recognition engine unit 11.
[0062]
After the processes in steps S9, S12, and S15, the process proceeds to step S16, and the application management unit 31 determines whether to end the robot control process. For example, the application management unit 31 determines whether or not the end button (not shown) is pressed by the user, and determines that the robot control process is to be ended when the end button is pressed.
[0063]
If it is determined in step S16 that the robot control process is not to be ended, the process returns to step S6 in FIG. 4 and repeats the process of recognizing the input voice. If the application management unit 31 determines in step S16 that the robot control process is to be ended (the end button has been pressed), the process proceeds to step S17, in which the common dictionary of the common dictionary unit 55 and the cluster of the unknown word acquisition unit 56 are created. The state is stored in a storage unit (not shown).
[0064]
Then, when there is an activated application unit 21, the application management unit 31 ends the application unit. At this time, the application unit 21 invalidates the corresponding task 71. Further, the application management unit 31 includes an application switching task 71. ₂ Is invalidated, and the speech recognition engine unit 11 executes the phoneme typewriter task 71. ₁ Is invalidated, and the application management unit 31 and the speech recognition engine unit 11 end the processing.
[0065]
Note that, in the above-described processing, the application section is executed by the name registration application section 21. ₁ , Chat application unit 21 ₂ , Voice commander application section 21 ₃ The above three cases have been described. However, if there is another application unit, if it is determined in step S13 that the voice commander application is not to be activated, the process does not return to step S6, and is similar to steps S7, S10, and S13. Then, it is determined whether or not to start another application, and the other application is started according to the determination result.
[0066]
Further, in the above-described processing, the end of the voice recognition is instructed by the user. However, the robot control system 1 may automatically determine, for example, end when no voice is input for a predetermined time.
[0067]
According to the above processing, the application switching task 71 ₂ Is valid even during the activation of each application unit 21. Therefore, even when the utterance "Start OO" is made during the activation of another application unit, the utterance is recognized and the corresponding Applications can be launched. For example, the voice commander application unit 21 ₃ When the user utters “Start chat” during the startup, the chat application unit 21 ₂ Can be started.
[0068]
In this case, terminate the running application section and then start the new application section, or place the running application section in a paused state and then start the new application section. Whether to restart the application unit or to start both in parallel is set in advance by a combination of the application units (it may be dynamically determined based on resource constraints such as memory).
[0069]
FIG. 6 shows the configuration of the task 71. The task 71 includes an acoustic model 111, a language model 112, a dictionary 113, a phoneme list 114, kana phoneme conversion rules 115, and search parameters 116.
[0070]
The acoustic model 111 stores a model representing acoustic features such as individual phonemes and syllables of the speech to be recognized. As the acoustic model, for example, HMM (Hidden Markov Model) can be used.
[0071]
The language model 112 describes information indicating how the words registered in the word dictionary of the dictionary 113 are linked (connected) (hereinafter, referred to as chain information as appropriate). Description methods include statistical word chain probability (n-gram), generation grammar, and finite state automan.
[0072]
The language model 112 includes, in addition to the chain information about the words, chain information about a category in which the words are classified from a specific viewpoint. For example, when the “category consisting of words representing user names is represented by a symbol“ _user name_ ”and the“ category consisting of words representing robot names ”is represented by a symbol“ _robot name_ ”, the language model 112 , “_User name_” and “_robot name_” (chains between categories, chains between categories and words stored in a dictionary in advance, etc.) are also described.
[0073]
Therefore, chain information can be obtained for words not included in the language model 112. For example, when acquiring the chain information of “SD” and “wa (particle)”, even if the language model 112 does not describe the chain information of “SD”, “SD” is “_ robot name”. If it is known that the robot belongs to the category represented by the symbol “_”, the chain information of “_robot name_” and “ha” is obtained instead, so that the chain of “SD” and “ha” are obtained. Information can be obtained.
[0074]
The category is not a classification based on the semantic attribute (“_robot name_”, “_user name_”, “_place name_”, “_store name_”, etc.), but a classification based on the part of speech (“_noun_ "," _Verb_ "," _particle_ ", etc.). Hereinafter, the notation “_... _” Represents a category name.
[0075]
The dictionary 113 includes a fixed word dictionary 131, a variable word dictionary 132, and a category table 133.
[0076]
The fixed word dictionary 131 stores pronunciations (phonological sequences), phonemes, and words for words that are not registered and deleted, that is, words (hereinafter, appropriately referred to as fixed words) set in advance in the robot control system 1. Various information such as a model describing a syllable chain relationship is described.
[0077]
The fixed word dictionary 131 describes, for each task 71, information about a dedicated word used by the application unit 21 corresponding to the task 71. The same applies to the above-described acoustic model 111 and language model 112, and also to a category table 133, phoneme list 114, kana phoneme conversion rules 115, and search parameters 116 described later.
[0078]
The variable word dictionary 132 describes various information such as a word to be registered and deleted, that is, a model that describes the pronunciation, phoneme, and syllable chain relation of the registered word. Is registered, the registered word is reflected. This reflection processing will be described later with reference to FIG. Deletion of words and change of pronunciation can be performed only for entries in the variable word dictionary 132. Note that the variable word dictionary 132 does not have to store anything.
[0079]
The category table 133 stores a table indicating a correspondence between a category included in the language model 112 and information on a word included in the category. When the task 71 assigns an ID unique to the category (category ID), the category table 133 also stores the correspondence between the symbol of the category and the ID. For example, when a category ID “4” is assigned to the category “_robot name_”, the category ID = 4 is also stored corresponding to “_robot name_”. Note that the category table 133 does not store anything when the language model 112 does not include a category.
[0080]
The phoneme list 114 is a list of phoneme symbols used in the task 71. The kana phoneme conversion rule 115 is a rule for converting a kana character string into a phoneme sequence. In this way, by storing the kana phoneme conversion rules 115 for each task, the common dictionary unit 55 can hold, as pronunciation information, a kana character string that is independent of the phoneme sequence.
[0081]
The search parameter 116 holds a parameter used when the matching unit 54 performs matching (search). The parameters include a value depending on the acoustic model 111, a value depending on the number of vocabulary words, a value depending on the type of the language model 112, and the like. However, parameters that do not depend on the task may be held in the recognition engine unit 11 in common.
[0082]
In the above description, all data is stored for each task. However, data used in common by a plurality of tasks can be shared between tasks to reduce the memory usage. For example, when the phoneme list 114 is common to all tasks, only one phoneme list 114 is prepared by the speech recognition engine unit 11, and each task may refer to it. In this case, it is sufficient to prepare only one kana phoneme conversion rule 115.
[0083]
In addition, two types of acoustic models 111 are prepared, one for a quiet environment (an acoustic model that provides a high recognition rate in a quiet environment) and one for a noise environment (an acoustic model that provides a reasonable recognition rate in a noisy environment). Either one may be referred to each time.
[0084]
For example, the name registration task 71 ₃ And chat task 71 ₄ Is assumed to be used in a quiet environment, so reference is made to the acoustic model 111 for the quiet environment, and the voice commander task 71 ₅ Is assumed to be used in a noisy environment (an environment in which the operation sound of the robot is loud), so that an acoustic model for a noisy environment can be referred to.
[0085]
FIG. 7 shows an example of the phoneme list 114 of FIG. In FIG. 7, one symbol represents (corresponds to) one phoneme. In the phoneme list 114 of FIG. 7, a vowel + colon (for example, “a:”) represents a long sound, and “N” represents a plucked sound (“n”). “Sp”, “silB”, “silE”, and “q” all represent silence, but “silence in speech”, “silence before speech”, “silence after speech”, and “prompting sound”. ("Tsu") ".
[0086]
FIG. 8 shows an example of the kana phoneme conversion rule 115 of FIG. According to the kana phoneme conversion rule 115 of FIG. 8, for example, a kana character string “SD” is converted into a phoneme sequence “e / s / u / d / i: / a: / r / u”. .
[0087]
Next, examples of the language model 112 and the dictionary 113 (FIG. 6) of each task are shown.
[0088]
FIG. 9 shows a phoneme typewriter task 71. ₁ 9 shows an example of the language model 112 (FIG. 6). In FIG. 9, the variable “$ SYLLABLE” on the first line means an arbitrary one of the kana notations because all kana notations are connected by “|” meaning “or”. .
[0089]
That is, here, the phonemic typewriter task 71 ₁ Is a task for speech recognition in units of syllables (syllables), and the language model 112 in FIG. 9 describes a chain rule that an arbitrary syllable can be arbitrarily connected in a BNF (Backus-Naur-Form) format. In the grammar. The language model 112 may use a statistical language model described later.
[0090]
FIG. 10 shows a phonemic typewriter task 71. ₁ 6 shows an example of the fixed word dictionary 131 (FIG. 6). The “symbol” is a character string for identifying a word, and can use, for example, kana notation. Entries with the same symbol are considered to be entries of the same word. The language model 112 is represented by using this symbol. Note that “<head>” and “<end>” are special symbols and represent “silence before utterance” and “silence after utterance”, respectively (the same applies to FIG. 11 and the like described later).
[0091]
Further, “transcription” indicates a notation of a word, and the character string output as a recognition result is this transcription. The “phonological sequence” is a representation of the pronunciation of a word as a phonological sequence.
[0092]
Phonemic Typewriter Task 71 ₁ Of the phonemic typewriter task 71 ₁ Nothing is stored because it is not assumed that words will be added to. Also, the phonemic typewriter task 71 ₁ Since the language model 112 does not include a category as shown in FIG. 9, nothing is stored in the category table 133.
[0093]
FIG. 11 shows an application switching task 71. ₂ 9 shows an example of the language model 112 (FIG. 6). The language model 112 in FIG. 11 is described in a BNF format grammar. The variable “$ APPLICATIONS” on the first line indicates that all application names (“chat”, “voice commander”, “name registration”, etc.) are connected by “|” which means “or”. Means one of the names.
[0094]
Also, in the variable “@UTTERANCE” on the second line, “[]” meaning “can be omitted” is added to each of “_robot name_” and “wo”, so “(robot name ) Start the application name ()). Here, the “robot name” indicates a word registered in the category of “_robot name_”.
[0095]
For example, when “SDR” is registered in “_robot name_”, utterances such as “SDR, voice commander (activate)” and “voice commander (activate)” are generated as shown in FIG. It is recognized by using 11 language models 112.
[0096]
By describing the language model 112 using the category name in this way, even if a newly registered word is included in the category described in the language model 112, Can recognize the utterance including the newly registered word by using the language model 112.
[0097]
FIG. 12 shows an application switching task 71. ₂ 6 shows an example of the fixed word dictionary 131 (FIG. 6). In the fixed word dictionary 131 of FIG. 12, transcriptions and phoneme sequences are described for symbols (“chat” and “voice commander” in FIG. 11) described in the grammar of the language model 112 of FIG. I have.
[0098]
FIG. 13 shows an application switching task 71. ₂ 6 shows an example of the category table 133 (FIG. 6). The category table 133 stores the type of category used for the language model 112 and information on words belonging to the category. When the language model 112 is as shown in FIG. ₂ Since the category of “_robot name_” is used for the language model 112, “_robot name_” is entered in the category table 133 as shown in FIG. In FIG. 13, the set of words belonging to the category of “_robot name_” is an empty set, indicating that there is no word belonging to “_robot name_”.
[0099]
As shown in FIG. 13, even when a category is entered in the category table 133, if there is no word belonging to the entry (in the case of an empty set), the variable word dictionary 132 stores The information of the word to which it belongs is not stored.
[0100]
FIG. 14 shows a name registration task 71. ₃ 9 shows an example of the language model 112 (FIG. 6). The language model 112 in FIG. 14 is described in a BNF format grammar. The variables "@UTTERNANCE" are "or [your name] is <OOV> [is] [it is called]" and "You [name] is <OOV> [it is]." Are connected by “|”, which means “name”, “is”, “says” and “says”, and “[]” is added to indicate “optional”. ing.
[0101]
Therefore, using the language model 112 of FIG. 14, “I (name) is <OOV> (is) (called)” or “You (name) is <OOV> (called)” Be recognized. Note that <OOV> is a symbol meaning “Out Of Vocabulary”, and means a phrase with an arbitrary pronunciation (a word not described in the fixed word dictionary 131).
[0102]
By using the symbol <OOV>, for example, utterances such as “My name is Taro” and “Your name is SDR” (“Taro” and “SDR” are described in the fixed word dictionary 131 14 are applied to the language model 112 in FIG. 14, “<head> My name is <OOV><end>” and “<head> Your name is <OOV>”. As a result, you can get the speech recognition results of "My name is Taro" and "Your name is SDR".
[0103]
FIG. 15 shows a name registration task 71. ₃ 6 shows an example of the fixed word dictionary 131 (FIG. 6). In the fixed word dictionary 131, transcriptions and phoneme sequences are described for the symbols described in the grammar of the language model 112 as shown in FIG.
[0104]
Name registration task 71 ₃ Here, the name registration task 71 ₃ Nothing is stored because it is not assumed that words will be added to. The name registration task 71 ₃ Since the language model 112 does not include a category as shown in FIG. 14, nothing is stored in the category table 133.
[0105]
FIG. 16 shows a chat task 71. ₄ 9 shows an example of the language model 112 (FIG. 6). Since the chat has many vocabulary and utterance variations, a statistical language model is used as the language model 112. The statistical language model is a model in which chain information of words is described with conditional probabilities. In the language model 112 of FIG. 16, a tri-line representing the arrangement of three words 1, 2, and 3, that is, the probability of three chains of words. Gram is used.
[0106]
In FIG. 16, “P (word 3 | word 1 word 2)” is a probability that “word 3” will appear next when “word 1” and “word 2” are arranged in the word string. Represents For example, when there is a sequence of “<head>“ _ robot name_ ””, the probability that “ha” appears next is “0.012”. This probability is obtained in advance by analyzing a text describing a large amount of chat. As the language model 112, besides the tri-gram, a bi-gram (probability of two chains), a uni-gram (probability of appearance of a word), and the like can be used as necessary.
[0107]
In the language model 112 of FIG. 16 as well, the grammar is described using categories in addition to words, as in the case of FIG. That is, in FIG. 16, “_robot name_” and “_place name_” mean the categories “_robot name_” and “_place name_”, and a tri-gram is described using these categories. Thus, when a word representing a robot name or a place name is registered in the variable word dictionary 132, the word is stored in the chat task 71. ₄ Can be recognized.
[0108]
FIG. 17 shows a chat task 71. ₄ Of the fixed word dictionary 131 of FIG. In the fixed word dictionary 131, transcription and phoneme sequences are described for symbols described in the grammar of the language model 112 as shown in FIG.
[0109]
FIG. 18 shows a chat task 71. ₄ Of the category table 133 of FIG. The category table 133 stores types of categories used for the language model 112 and information on words belonging to the categories. When the language model 112 is as shown in FIG. ₄ Since two categories “_robot name_” and “_location name_” are used in the language model 112 of “.”, The category table 133 includes “_robot name_” as shown in FIG. And "_place name_" are entered. In FIG. 18, words belonging to the categories “_robot name_” and “_place name_” indicate that there is no word yet.
[0110]
FIG. 19 shows a task 71 for voice commander. ₅ 9 shows an example of the language model 112 (FIG. 6). The language model 112 in FIG. 19 is described in a BNF-format grammar. The variable “@NUMBER” on the first line indicates that any one of the numbers (“1”, “2”, “3”, etc.) Means one.
[0111]
The variable “$ DIRECTION” on the second line is connected with “|” meaning “or” in the directions (“front”, “back”, “right”, “left”, etc.). Means any one of The variable “UTTERANCE” in the third line is obtained by adding “advance” to “_robot name_”, “$ DIRECTION”, and “$ NUMBER step”. "[]" Meaning "can be omitted" is added to "_robot name_", "@DIRECTION", and "@NUMBER step".
[0112]
Therefore, in the language model 112 of FIG. 19, for example, a voice such as “three steps ahead of (robot name)” is recognized.
[0113]
FIG. 20 shows a task 71 for a voice commander. ₅ Of the fixed word dictionary 131 of FIG. In the fixed word dictionary 131, transcriptions and phoneme sequences are described for symbols described in the grammar of the language model 112 as shown in FIG.
[0114]
Note that the symbols of “1” and “step” overlap, which means that “1” and “step” have two pronunciations (“ichi” and “it”, “ho” and “po”, respectively). ]). As a result, for example, utterances with different pronunciations such as “Ichiho” and “Ippo” can be recognized as the same “one step”.
[0115]
When the language model 112 is as shown in FIG. ₅ Since only the category “_robot name_” is used in the language model 112 of ₅ The category table 133 of the application switching task 71 shown in FIG. ₂ Is the same as the category table 133 of FIG. In the state where the word belonging to “_robot name_” has not been uttered yet, the voice commander task 71 ₅ Nothing is stored in the variable word dictionary 132.
[0116]
Next, the name registration application unit 21 ₁ However, the name registration process performed in step S9 in FIG. 5 will be described in detail with reference to the flowchart in FIG. This processing is performed by the user's utterance, ₁ Triggered when is invoked. Before this processing is started, the user may input a name by voice input mode of inputting a name by voice or kana input by a keyboard or the like as a name registration mode for registering a name by a mode switching button (not shown), for example. One of the kana input modes for input is selected in advance.
[0117]
In step S41, the name registration application unit 21 ₁ Is a name registration task 71 of the speech recognition engine unit 11. ₃ And enable this name registration task 71 ₃ To be able to recognize voice.
[0118]
After the process in step S41, the process proceeds to step S42, where the name registration application unit 21 ₁ Determines whether or not the name registration mode is the voice input mode. If it is determined that the name registration mode is the voice input mode, the process proceeds to step S43, in which the matching unit 54 performs a name recognition process, and then proceeds to step S44. (Or, if the user utters in step S42, it is determined that "the name has been input by voice," and the process proceeds to step S43. If the kana input button (not shown) has been pressed, the user has input the And proceeds to step S46.) Details of this name recognition processing will be described later with reference to FIG.
[0119]
In step S44, the name registration application unit 21 ₁ Determines whether the speech recognition result (recognized name) of the name obtained by performing the name recognition process in step S43 by the matching unit 54 is correct. This determination is made based on, for example, whether the recognition result is uttered to the user and the user operates an OK button (not shown).
[0120]
If it is determined in step S44 that the name speech recognition result is incorrect, the user is prompted to speak again, and the process returns to step S43 to perform name recognition processing again. If it is determined in step S44 that the recognition result is correct, the process proceeds to step S47.
[0121]
On the other hand, in step S42, the name registration application unit 21 ₁ Determines that the name registration mode is not the voice input mode, proceeds to step S45, and determines whether the name registration mode is the kana input mode.
[0122]
If it is determined in step S45 that the name registration mode is not the kana input mode, the user has not selected the name registration mode, so the user waits until the name registration mode is selected. After waiting, the process returns to step S42.
[0123]
If it is determined in step S45 that the name registration mode is the kana input mode, the process proceeds to step S46, where the name registration application unit 21 ₁ Acquires the kana sequence of the name input by the user and the category of the name.
[0124]
As a method of inputting the kana sequence, for example, a method in which a user temporarily connects a keyboard to input kana characters, a method of inputting using various switches of the robot, a method of displaying paper on which characters are written to the robot, or the like. (For example, refer to Japanese Patent Application No. 2001-135423), a method of connecting a robot to a personal computer by a wireless LAN (Local Area Network) or the like, and transferring the robot from the personal computer to the robot, There is a method of downloading to a robot. Also, in the method of recognizing characters by showing a paper or the like on which characters are written to a robot, instead of inputting kana characters, a character string containing kana and kanji characters is input, and a name registration application unit 21 is input. ₁ May be converted to kana strings (see Japanese Patent Application No. 2001-135423).
[0125]
Further, instead of the user inputting the kana string of the name, an entry to which the kana character of the name is added is given to the common dictionary of the common dictionary unit 55 in advance, and the name registration application unit 21 is provided. ₁ May refer to the common dictionary unit 55 to obtain a kana sequence of names.
[0126]
After the processing in step S44 or S46, the process proceeds to step S47, where the name registration application unit 21 ₁ Determines the category of the name to be registered. When the name registration mode is the kana input mode, the name registration application unit 21 ₁ Determines the category (input by the user) acquired in step S46 as the category of the name to be registered.
[0127]
That is, in the kana input mode, in step S46, the user is requested to input a category of the name in addition to the name, and the category of the name input by the user is determined as a category of the name to be registered. On the other hand, when the name registration mode is the voice input mode, the name registration application unit 21 ₁ Is determined by estimating the category of the name obtained in the name recognition processing in step S43.
[0128]
For example, if the recognition result supplied from the voice recognition engine unit 11 starts with “Kimi”, the category to which the registered name belongs is assumed to be “_robot name_”, and if it starts with “I”, It is assumed that the category to which the registered name belongs is “_user name_”. Further, various category estimation methods disclosed in Japanese Patent Application No. 2001-382579 previously proposed by the present applicant can also be used.
[0129]
After the process of step S47, the process proceeds to step S48, where the name registration application unit 21 ₁ Controls the matching unit 54 to enter the pronunciation information and category of the name to be registered in the common dictionary of the common dictionary unit 55, and proceeds to step S49. In step S49, the name registration application unit 21 ₁ Controls the matching unit 54 to store the contents of the common dictionary in the chat task 71. ₄ , Voice commander task 71 ₅ , ..., other tasks 71 _N To reflect. The details of this reflection will be described later with reference to the flowchart in FIG.
[0130]
In this way, by reflecting the name registered in the common dictionary in another task, the registered name can be recognized in another task.
[0131]
After the process of step S49, the process proceeds to step S50, where the name registration application unit 21 ₁ Determines whether to end the name registration process. This determination is made, for example, by speaking to the user a question as to whether or not to end, and whether or not the user has operated (pressed) an OK button (not shown). If it is determined in step S49 that the name registration application is not to be terminated (for example, the OK button has not been pressed), the process returns to step S42 to perform processing for registering another name.
[0132]
If it is determined in step S50 that the name registration process is to be ended (for example, the OK button has been pressed), the process proceeds to step S51, and the name registration application unit 21 ₁ Is the name registration task 71 ₃ Is invalidated, and the process proceeds to step S52. In step S52, the name registration application unit 21 ₁ Ends the processing.
[0133]
FIG. 22 is a flowchart illustrating a name recognition process performed by the matching unit 54 of FIG. 2 in step S43 of FIG.
[0134]
In step S61, the matching unit 54 determines whether or not a voice has been input to the microphone 51. If it is determined that no voice has been input, the matching unit 54 waits until a voice is input. If it is determined in step S61 that a voice has been input, the process proceeds to step S62. The voice input here may be a normal conversation such as "My name is Taro" or "Your name is SDR", and the user is aware of the name registration and "Taro"" You don't have to enter the name "SDR" alone.
[0135]
In step S62, the matching unit 54 recognizes the voice and extracts the name. For example, if the utterance “Your name is SDR” is made, a name registration task 71 having a language model 112 as shown in FIG. 14 and a fixed word dictionary 131 as shown in FIG. ₃ , The matching unit 54 generates a recognition result that “Your name is <OOV><End>”, for example. Also, the matching unit 54 obtains information indicating which section of the utterance the <OOV> is (from what second to what second of the first utterance).
[0136]
Further, the matching unit 54 performs a phoneme typewriter task 71 having a language model 112 as shown in FIG. 9 and a fixed word dictionary 131 as shown in FIG. ₁ With reference to, for example, “k / i / m / i / n / o / n / a / m / a / e / w / a / e / s / u / d / i: / a: / r / A phonemic sequence “u / d / a / y / o” and a kana sequence “Kimino Nama Eva Sd R Dayo” are obtained.
[0137]
Then, based on information indicating which section of the utterance the <OOV> is, the matching unit 54 determines, based on the obtained phoneme sequence and kana sequence, a section corresponding to <OOV>, that is, a phoneme sequence of the name section. The kana sequence is cut out to obtain a phoneme sequence "e / s / u / d / i: / a: / r / u" and a kana sequence "ESD". Further, the matching unit 54 also obtains audio data of the same section. The details of the process of extracting the name are disclosed in Japanese Patent Application No. 2001-382579 previously proposed by the present applicant.
[0138]
After the process in step S62, the process proceeds to step S63, in which the matching unit 54 supplies the phoneme sequence, kana sequence, and voice data of the name extracted in the process in step S62 to the unknown word acquisition unit 56, and performs clustering. The details of the clustering are disclosed in Japanese Patent Application No. 2001-097845 previously proposed by the present applicant. As a result of this clustering, each cluster of the unknown word acquisition unit 56 has a representative phoneme sequence and a kana sequence.
[0139]
After the processing in step S63, the process proceeds to step S64, in which the recognition result of the voice recognized in step S62 (for example, the kana sequence "Kimino Nama Eva Sd R Dayo") is stored in the name registration application section 21. ₁ To supply.
[0140]
FIG. 23 shows an example of the feature space clustered by the unknown word acquisition unit 56 in the process of step S63 in FIG. Note that FIG. 23 shows a feature space defined by two feature amounts (feature parameters) 1 and 2 to avoid complicating the drawing (the same applies to FIG. 3 described above). In FIG. 23, in the feature space, four names “arara”, “sana”, “tokyo”, and “taro” are clustered.
[0141]
That is, in FIG. 23, in the feature space, four clusters of an “arara” cluster 151, a “saw” cluster 152, a “tokyo” cluster 153, and a “tarou” cluster 154 are formed, Each cluster has a representative phoneme series (in the example of FIG. 23, “a / r / a / r / a”, “s / a / n / i:”, “t / o: / ky / o”). : "," T / a / r / o / u "), representative kana notation (" Alara "," Sunny "," Tokyo "," Taro "in the example of FIG. 23), and ID (FIG. In the example of 23, “1”, “2”, “3”, “5”) are added.
[0142]
FIG. 24 shows an example of a common dictionary of the common dictionary unit 55 in which word information has been entered in step S48 of FIG. In FIG. 24, the entry on the first line is that the pronunciation is input in a kana sequence, the pronunciation is a character string of "SD", and indicates that the category is input as "_robot name_". .
[0143]
The entry on the second line is a representative kana notation in which the pronunciation is input in speech, and the kana notation and the phonological sequence of the pronunciation are added to the cluster whose ID of the unknown word acquisition unit 56 is “5” (FIG. 23). In the example of FIG. 23, “tarrow”) and the phoneme sequence (“t / a / r / o:” in the example of FIG. 23). The category of the entry on the second line is determined by the name registration application unit 21 in step S47 in FIG. 21 and is “_user name_”. For example, when the user utters “My name is Taro”, an entry like the second line is configured in the common dictionary unit 55.
[0144]
Similarly, in the entries on the third and fourth lines, pronunciations are input in kana columns, and the pronunciations are character strings "Sunitarou" and "Kitashinagawa", respectively, and the category is "_user name_". This indicates that “_place name_” has been input. In the entry on the fifth line, the pronunciation is input in speech, and the kana notation and the phoneme sequence of the pronunciation are represented by a representative kana notation (ID) of the unknown word acquisition unit 56 added to the cluster with the ID “3”. In the case of the example of FIG. 23, “Tokyo”) and the phoneme sequence (“t / o: / ky / o:” in the example of FIG. 23) are shown. Further, the category of the entry on the fifth line is determined to be “_place name_” by the name registration application unit 21.
[0145]
In the common dictionary, for a word whose pronunciation is entered in a kana sequence, a set of a kana sequence and a category representing the pronunciation of the word is registered in one entry, and for a word whose pronunciation is spoken, A pair of an ID and a category representing the cluster of the word is registered in one entry.
[0146]
FIG. 25 is a flowchart illustrating a process in which the matching unit 54 reflects the contents of the common dictionary unit 55 to the task in the process of step S49 in FIG. This process is performed for each enabled task.
[0147]
In step S81, the matching unit 54 initializes the variable word dictionary 132 and the category table 133 in the task 71 (FIG. 6). That is, the variable word dictionary 132 has no entry, and the category table 133 has a state in which no word belongs to each category.
[0148]
After the process in step S81, the process proceeds to step S82, where the matching unit 54 reflects the contents of the common dictionary unit 55 in the variable word dictionary 132 and the category table 133.
[0149]
That is, the matching unit 54 selects, from the common dictionary of the common dictionary unit 55, a category common (same) to the category entered in the category table 133, and selects the category and the cluster ID corresponding to the category. Or, get kana pronunciation (kana sequence). Further, when acquiring the cluster ID from the common dictionary, the matching unit 54 acquires a kana sequence corresponding to the cluster ID from the unknown word acquisition unit 56.
[0150]
When the matching unit 54 obtains the kana sequence of the word belonging to the selected category from the common dictionary of the common dictionary unit 55 as described above, the matching unit enters the kana sequence into the variable word dictionary 132. Further, the matching unit 54 enters the information of the word represented by the kana sequence acquired from the common dictionary into the corresponding category of the category table 133.
[0151]
According to the above-described processing, in each task, the contents of the common dictionary are reflected after the variable word dictionary 132 is initialized. That is, the variable word dictionary 132 is constructed or reconstructed based on the contents of the common dictionary. Therefore, it is possible to easily maintain consistency in each task as compared with a method of deleting or changing a specific entry in the dictionary.
[0152]
Further, according to the above-described processing, the word registered in the voice is acquired in the variable word dictionary 132 because the latest pronunciation information at that time is acquired from the unknown word acquisition unit 56 every time the word is reflected in each task. Thereafter, the pronunciation information is updated simply by supplying the speech data to the unknown word acquisition unit 56, and the matching unit 54 can recognize the speech by referring to the latest pronunciation information at that time.
[0153]
FIG. 26 is a block diagram illustrating the reflection processing of FIG. When the kana sequence is described in the common dictionary of the common dictionary unit 55 corresponding to the category, the kana sequence is registered in the variable word dictionary 132, and the category table 133 has the same category as the category of the common dictionary. Information of words represented by kana strings in the common dictionary is registered.
[0154]
On the other hand, if the common dictionary describes a cluster ID corresponding to the category, the unknown word acquisition unit 56 is referred to, and the representative kana sequence and the representative phoneme sequence corresponding to the cluster ID are registered in the variable word dictionary 132. Then, the information of the word represented by the cluster ID of the common dictionary is registered in the same category as the category of the common dictionary in the category table 133. In the speech recognition processing described later, both the fixed word dictionary 131 and the variable word dictionary 132 are used.
[0155]
FIG. 27 shows an application switching task 71 reflecting the contents of the common dictionary unit 55 shown in FIG. ₂ Is an example of the variable word dictionary 132. Application switching task 71 ₂ When the category table 133 shown in FIG. 13 is as shown in FIG. 13, the category common to the common dictionary in FIG. 24 is “_robot name_”. A kana pronunciation of "SD" corresponding to name_ "is acquired.
[0156]
Then, as shown in FIG. 27, the matching unit 54 enters the kana pronunciation “SDR” acquired from the common dictionary in FIG. 24 into the transcription of the variable word dictionary 132. Further, the matching unit 54 adds “e / s / u /” corresponding to the kana pronunciation “SD” to the phoneme sequence corresponding to the transcription “SD” based on the kana phoneme conversion rule 115 (FIG. 8). d / i: / a: / r / u ".
[0157]
In addition, the matching unit 54 registers “OOV00001” as a symbol of the word represented by the transcription “SDR”. Here, the symbol is “OOV00001”, which means “OOV” + serial number, but the symbol may be any character string that can uniquely identify the word. That is, as the symbol, for example, "_robot name _ :: OOV00001" or the like can be used by adding a category name to the head.
[0158]
FIG. 28 shows an application switching task 71 in which the contents of the common dictionary of FIG. 24 are reflected. ₂ Of the category table 133 of FIG. As shown in FIG. 27, when the contents of the common dictionary of FIG. 24 are reflected in the variable word dictionary 132, the contents of the category table 133 are stored in the category of “_robot name_” shown in FIG. Are registered, the symbol “OOV00001” of the word belonging to the category “_robot name_” registered in the variable word dictionary 132 of FIG. 27 is entered.
[0159]
Next, a process in which the matching unit 54 deletes or changes a word registered in the common dictionary of the common dictionary unit 55 in step S48 of FIG. 21 will be described with reference to a flowchart of FIG. The process of deleting or changing words in the common dictionary is started, for example, when there is a command from the name registration application unit 21 or when it becomes necessary to delete unnecessary words from registered words due to memory restrictions. Is done.
[0160]
In addition, the process of deleting or changing the words in the common dictionary includes, for example, changing the ID assigned to the cluster by deleting the cluster in the unknown word acquisition unit 56 or dividing or merging the cluster. When it is necessary to match the ID assigned to the cluster of the unknown word acquisition unit 56 with the ID described in the common dictionary (the cluster ID described in FIG. 24), the ID described in the common dictionary is changed to This is done to rewrite.
[0161]
Further, when the common dictionary is deleted or changed, when a certain category is no longer used by all the tasks described in the language model 112, the category information is deleted from the common dictionary, and the common dictionary is slimmed. This is done in order to achieve
[0162]
If the unknown word acquisition unit 56 changes the representative phoneme sequence and the kana sequence of the cluster, the changes are reflected in the common dictionary by the reflection processing in FIG. 25, so that the words are deleted or changed. It is not necessary to perform the processing (hereinafter, appropriately referred to as change deletion processing).
[0163]
In step S101, the matching unit 54 determines a word to be subjected to change deletion processing from the common dictionary, and proceeds to step S102. The target word may be determined by the user using a button (not shown), or may be determined and determined by the matching unit 54.
[0164]
In step S102, the matching unit 54 determines whether or not to delete the word targeted for the change deletion process. If it is determined that the word is to be deleted, the process proceeds to step S103. In step S103, the matching unit 54 deletes the entry of the word targeted for the change deletion process from the common dictionary. Deletion means deleting entries specified by a category and pronunciation information, deleting entries of a specific category at once, or deleting entries having specific pronunciation information (kana sequence or cluster ID) at once. Means to do.
[0165]
On the other hand, if the matching unit 54 determines in step S102 that the word to be subjected to the change deletion process is not to be deleted, the process proceeds to step S104, where it is determined whether the word is to be changed, and it is determined that the word is not to be changed. In this case, the process returns to step S102 and waits until it is determined that the change or the deletion is performed.
[0166]
When the matching unit 54 determines in step S104 to change the word to be subjected to the change deletion process, the process proceeds to step S105, and changes the entry of the word to be subjected to the change deletion process in the common dictionary.
[0167]
For example, when the cluster of the unknown word acquiring unit 56 is divided or merged and the ID number of the cluster changes, the matching unit 54 performs the matching with the unknown word acquiring unit 56 by using the cluster ID of the common dictionary. To change. Further, for example, when the user wants to later correct the kana sequence input at the time of registration, the matching unit 54 uses the name registration application unit 21. ₁ , The kana pronunciation of the word to be a target of the common dictionary (the word entered in the common dictionary in step S48 of FIG. 21) is converted to the kana sequence input by the user after determining the word to be a target of the common dictionary. change.
[0168]
After the processing of step S103 or the processing of step S105, the process proceeds to step S106, where the matching unit 54 performs the reflection processing of FIG. 25, and reflects the contents of the common dictionary to each task.
[0169]
As described above, when a word in the common dictionary is deleted or changed, the changed content is reflected in each task, so that the consistency of the registered words in each application unit can be maintained.
[0170]
FIGS. 30 and 31 show an example in which the matching unit 54 changes the entry of a word in the common dictionary in the process of step S105 in FIG. For example, when the cluster with the ID “5” of the unknown word acquisition unit 56 is divided into the cluster with the ID “8” and the cluster with the ID “9”, the matching unit 54 shows the common dictionary in FIG. 30A. The state is changed from the state shown in FIG. 30B to the state shown in FIG. 30B.
[0171]
That is, the matching unit 54 deletes the entry with the cluster ID “5” (the entry on the first line in FIG. 30A) of the common dictionary unit 55, and “_user name_” registered in the deleted entry. Register two entries of the category. Further, the matching unit 54 describes the cluster ID numbers “8” and “9” in the two new entries, respectively (entries on the first and second lines in FIG. 30B).
[0172]
Further, for example, when the cluster with the cluster ID “5” and the cluster with the ID “3” of the unknown word acquisition unit 56 are merged to generate a new cluster with the ID “10”, the matching unit 54 Then, the common dictionary is changed from the state shown in FIG. 31A to the state shown in FIG. 31B.
[0173]
That is, the matching unit 54 changes the cluster ID of the entries having the cluster IDs “5” and “3” of the common dictionary (all the entries in FIG. 31A) to “10”. The two entries of the category “_” and the corresponding cluster ID number “10” are made one (for example, one is deleted) (FIG. 31B).
[0174]
Next, the chat processing in step S12 in FIG. 5 will be described in detail with reference to the flowchart in FIG.
[0175]
In step S121, the chat application unit 21 ₂ Is a chat task 71 ₄ Is enabled, and the process proceeds to step S122. In step S122, the chat application unit 21 ₂ Controls the matching unit 54 to perform the reflection processing as shown in FIG. 25, and stores the contents of the common dictionary of the common dictionary unit 55 into the chat task 71. ₄ (The variable word dictionary 132 and the category table 133). Therefore, the chat task 71 ₄ Can obtain words registered, changed, and deleted in the common dictionary while being invalid.
[0176]
After the process of step S122, the process proceeds to step S123, and the chat application unit 21 ₂ Controls the voice recognition engine unit 11 to perform voice recognition processing, and proceeds to step S124. Details of the voice recognition processing will be described later with reference to FIG.
[0177]
In step S124, the chat application unit 21 ₂ Acquires a recognition result from the speech recognition engine unit 11 and generates a response to the recognition result. That is, the robot responds to the utterance from the user. For example, if the utterance from the user is “When did the SDR (robot name) wake up?”, The chat application unit 21 ₂ Generates a response for the time the robot was awakened (activated) (eg, “7:00”) and causes the robot to speak.
[0178]
After the process of step S124, the process proceeds to step S125, and the chat application unit 21 ₂ Determines whether to end the processing. This determination is made, for example, by the chat application unit 21. ₂ However, the user makes an utterance “End?” To the user, and determines whether the user has operated (pressed) (pressed) the OK button (not shown).
[0179]
If it is determined in step S125 that the process is not to be ended, the process returns to step S123, and the same process is repeated. That is, the robot continues the chat with the user.
[0180]
If it is determined in step S125 that the process is to be ended, the process proceeds to step S126, and the chat application unit 21 ₂ Is a chat task 71 ₄ Is invalidated, and the process proceeds to step S127. In step S127, the chat application unit 21 ₂ Ends the processing.
[0181]
In the above-described processing, the chat application unit 21 each time the user speaks once. ₂ Generated the response, but the robot may prompt the user to speak by spontaneously speaking.
[0182]
In the process of FIG. 32, the chat application unit 21 ₂ Has been described, the voice commander application unit 21 ₃ Voice commander processing,..., And other application units 21 _M Is similarly performed. However, in step S124, processing based on the speech recognition result by the speech recognition engine unit 11 is performed according to the application unit 21.
[0183]
FIG. 33 shows the contents of the common dictionary of the common dictionary unit 55 shown in FIG. ₄ In the variable word dictionary 132 shown in FIG.
[0184]
Chat task 71 ₄ When the category table 133 shown in FIG. 18 is as shown in FIG. 18, the categories common to the common dictionary in FIG. 24 are “_robot name_” and “_place name_”, so that the matching unit 54 The first entry in FIG. 24 is acquired as a common dictionary entry corresponding to “name_”, and the fourth and fifth entries in FIG. 24 are acquired as entries corresponding to “_place name_”. Further, from the first entry, the pronunciation of kana "ESD" is obtained, from the fourth entry, the pronunciation of kana "Kitashinagawa" is obtained, and from the fifth entry, the cluster ID number "3" is obtained.
[0185]
Then, as shown in FIG. 33, the matching unit 54 enters “SDR” and “Kitashinagawa” into the transcription of the variable word dictionary 132. Further, the matching unit 54 adds “e / s / u / d / i:” to the phoneme sequence of the variable word dictionary 132 based on the kana phoneme conversion rule 115 (FIG. 8) in correspondence with the transcription “SDR”. "/ A: / r / u" and "k / i / t / a / sh / i / n / a / g / a / w / a" corresponding to the transcription "Kitashinagawa".
[0186]
Further, the matching unit 54 extracts the cluster with the cluster ID “3” from the unknown word acquisition unit 56, and acquires a representative phoneme sequence and a kana sequence. For example, when the unknown word acquisition unit 56 is in a state as shown in FIG. 23, the matching unit 54 starts the cluster 153 with the cluster ID “3” and outputs the phoneme sequence “t / o: / ky / o:” and “ To get the kana sequence "Tokyo". Then, as shown in FIG. 33, the matching unit enters the acquired phoneme sequence “t / o: / ky / o:” and the kana sequence “Tokyo” into the phoneme sequence and the transcription of the variable word dictionary 132, respectively. I do.
[0187]
Further, the matching unit 54 transcribes “OOV00001” as the symbol of the word represented by the transcription “SDR” and “OOV00002” as the symbol of the word represented by the transcription “Kitashinagawa”. “OOV00003” is registered as a symbol of the word represented by the option “Tokyo”.
[0188]
In this case, the phonemic typewriter task 71 ₁ And chat task 71 ₄ Of the phoneme typewriter task 71 ₁ The representative phoneme sequence of the cluster obtained by using ₄ Is registered in the variable word dictionary 132, but the phonetic typewriter task 71 for kana ₁ And chat task 71 ₄ If the kana phoneme sequence rules 115 of the cluster are different, the matching unit 54 acquires a representative kana sequence of the cluster from the unknown word acquisition unit 56, and the chat task 71 ₄ Based on the kana phoneme sequence rule 115, the phoneme sequence of the variable word dictionary 132 is described.
[0189]
FIG. 34 shows that the contents of the common dictionary of FIG. ₄ Is reflected in the category table 133 of FIG. In the category table 133, for the category of “_robot name_”, the words belonging to the category “_robot name_” (the words whose transcription is “SD” (FIG. 33)) are stored in the variable word dictionary 132. The symbol “OOV00001” is entered. Further, for the category of “_place name_” in the category table 133, the words belonging to the category “_place name_” (the words of the transcriptions “Kitashinagawa” and “Tokyo” (FIG. 33)) are stored in the variable word dictionary 132. The registered symbols “OOV00002” and “OOV00003” are entered.
[0190]
Next, the speech recognition processing performed by the speech recognition engine unit 11 in FIG. 2 in the processing in step S123 in FIG. 32 will be described in detail with reference to the flowchart in FIG. This processing is started when a voice is input from the user to the microphone 51, and the application switching task 71 is started. ₂ , Chat task 71 ₄ , Voice commander task 71 ₅ , ..., other tasks 71 _N Of these tasks, this is performed for each enabled task.
[0191]
The audio signal generated by the microphone 51 is converted into audio data as a digital signal by the AD conversion unit 52 in step S141, and is supplied to the feature amount extraction unit 53. After the process in step S141, the process proceeds to step S142, in which the feature amount extraction unit 53 extracts a feature amount such as a mel cepstrum from the supplied audio signal, and proceeds to step S143.
[0192]
In step S143, the matching unit 54 connects some of the words represented by the symbols of the fixed word dictionary 131 and the variable word dictionary 132, generates a word string, and calculates an acoustic score. The acoustic score indicates how close (acoustically) the word string that is a candidate for the speech recognition result and the input speech are (acoustically) as sounds.
[0193]
After the process of step S143, the process proceeds to step S144, where the matching unit 54 selects a predetermined number of word strings having a high acoustic score based on the acoustic score calculated in step S143, and proceeds to step S145.
[0194]
In step S145, the matching unit 54 calculates the language score of each word string selected in step S144 using the language model 112, and proceeds to step S146. For example, when a grammar or a finite state automan is used as the language model 112, when the word string can be accepted by the language model 112, the language score is “1”, and when it cannot be accepted. , The language score is “0”.
[0195]
The matching unit 54 may leave the word string selected in step S144 when it can be accepted, and may delete the word string selected in step S144 when it cannot be accepted.
[0196]
When a statistical language model is used as the language model 112, the generation probability of the word string is used as a language score. The details of the method of obtaining the language score are disclosed in Japanese Patent Application No. 2001-382579 previously proposed by the present applicant.
[0197]
For example, the voice commander application unit 21 ₃ When the voice recognition processing is performed in the voice commander processing of the above, when the matching unit 54 selects the word string “<head> OOV00001 before moving forward <end>” in the processing of step S144, the language score is determined by the word string “< The head> OOV00001 advance to <end> ”can be accepted by the grammar language model 112 shown in FIG.
[0198]
That is, the matching unit 54 recognizes that the category of the symbol “OOV00001” is “_robot name_” with reference to the category table 133 (FIG. 28), and the word string “<head” obtained in step S144. > OOV00001 advance <end> is converted to a word string “<head> _robotic name_advance <end>” using a category name, and accepted by the language model 112 shown in FIG. Determine that you can.
[0199]
On the other hand, for example, when the word sequence “<head> OOV00001 and before <end>” is selected in step S144, the matching unit 54 refers to the category table 133 (FIG. 28) and refers to the category of the symbol “OOV00001”. Is recognized as "_robot name_", and the word string "<head> OOV00001 and proceeding to <end>" obtained in step S144 is replaced with the word string "<head> _robot name" using the category name. _ To "before <end>", it is determined that the language model 112 shown in FIG. 19 cannot accept the word sequence, and the language score of this word string is set to "0".
[0200]
In step S146, the matching unit 54 sorts the word strings by integrating the acoustic score calculated in step S143 and the language score calculated in step S145, and for example, sorts the word strings having the largest integrated score. Is determined as a recognition result.
[0201]
As a result, a word string most suitable acoustically and linguistically is determined as a recognition result.
[0202]
After the process in step S146, the process proceeds to step S147, and the matching unit 54 determines whether the recognition result includes a word registered as a voice (a word clustered in the unknown word acquisition unit 56).
[0203]
If it is determined in step S147 that the word registered in the voice is included in the recognition result, the process proceeds to step S148, where the matching unit 54 supplies the word to the unknown word acquiring unit 56, and 56 performs re-clustering. Then, the process proceeds to step S149.
[0204]
For example, in step S144, if the word string “<top> today went to Tokyo <end>” including the place name (unknown word) “Tokyo” is obtained, the matching unit 54 is an unknown word. "Tokyo" voice data, phonemic typewriter task 71 ₁ Is supplied to the unknown word acquisition unit 56 with the phoneme sequence (for example, “t / o: / ky / o:”) and the kana sequence (for example, “Tokyo”) recognized with reference to. Then, the unknown word acquisition unit 56 performs re-clustering.
[0205]
As a result, the amount of voice data supplied to the unknown word acquisition unit 56 increases, and the representative phoneme sequence and the representative kana sequence of each cluster may be updated to correct values. However, as a side effect, even after a correct phonemic kana sequence / kana sequence is obtained, the value may be changed to an incorrect value by re-clustering. In order to prevent such side effects, if there is an instruction from the user, the pronunciation can be fixed by describing the kana sequence at that time in an entry of the common dictionary. For example, in FIG. 23, when the kana sequence of the cluster with the ID = 3 is pronounced as “Tokyo”, the part described as “Cluster ID = 3” in the common dictionary in FIG. 24 is changed to “Kana pronunciation: Tokyo”. (The fifth entry is to be rewritten). By doing so, even if the kana sequence of the cluster with ID = 3 changes to a value other than “Tokyo”, the pronunciation of the fifth entry in the common dictionary is fixed at “Tokyo”.
[0206]
On the other hand, if the matching unit 54 determines in step S147 that the word registered in the voice is not included in the voice recognition result, the process skips step S148 and proceeds to step S149.
[0207]
In step S149, the matching unit 54 supplies the recognition result determined in the process of step S146 to the application unit 21 corresponding to the task.
[0208]
Here, the chat application unit 21 ₂ In the chat processing of FIG. 36, an expression for calculating a language score when the matching unit 54 selects the word string “<head> OOV00001 occurred at <end>” in step S144 of FIG. Show.
[0209]
As shown in equation (1), the language score “Score (<head> OOV00001 occurred at <end>)” is a word string “<head> OOV00001 occurred at <end>” as shown in equation (1). Probability.
[0210]
The value of the language score “Score (<head> OOV00001 is what occurred <end>)” is exactly “P (<head>) P (OOV00001 | <head”) as shown in equation (2). >) P (was | <head> OOV00001) P (when | <head> OOV00001 is) P (is | <head> OOV00001 is what time) P (was || <head> OOV00001 is when) P (| < The head> OOV00001 is obtained at what time occurred P (<end> | <head> OOV00001 was generated at what time). As shown in FIG. 16, the language model 112 uses tri-gram. Therefore, the condition parts "<head> OOV00001 is", "<head> OOV00001 is what time", "<head> OOV00001 is what time", "<head>OOV0" 001 happened at what time "and"<head> OOV00001 happened at what time "are up to two words immediately before" OOV00001 is "," was what time "," what time "," was happened ", and It is approximated by conditional probabilities each limited to "was happened" (equation (3)).
[0211]
The conditional probability is obtained by referring to the language model 112 (FIG. 16). However, since the language model 112 does not include the symbol “OOV00001”, the matching unit 54 refers to the category table 133 in FIG. Then, it recognizes that the category of the word represented by the symbol “OOV00001” is “_robot name_”, and converts “OOV00001” to “_robot name_”.
[0212]
That is, as shown in Expression (4), “P (OOV00001 | <head>)” is changed to “P (_robot name_ | <head>) P (OOV00001 | _robot name)” and “P (OOV00001 | _robot name)” (_Robot name_ | <head>) "/ N". N represents the number of words belonging to the category of “_robot name_” in the category table 133.
[0213]
That is, when the probability is described in the form of P (X | Y), when the word X is a word belonging to the category C, P (C | Y) is obtained from the language model 112, and the value is represented by P (X | C) (the probability that the word X is generated from the category C). Assuming that all words belonging to category C are generated with equal probability, if there are N words belonging to category C, P (X | C) can be approximated as 1 / N.
[0214]
In FIG. 34, since only the word represented by the symbol “OOV00001” belongs to the category “_robot name_”, N becomes “1”. Therefore, as shown in equation (5), “P (|| <head> OOV00001)” becomes “P (= | <head> _robot name_)”. In addition, “P (what | OOV00001)” becomes “P (what | _ robot name _)” as shown in Expression (6).
[0215]
Thus, a language score can be calculated for a word string including a variable word, and the variable word can appear in the recognition result.
[0216]
In the above example, the activation of the application unit 21 and the validity of the task 71, and the termination of the application unit 21 and the invalidation of the task 71 are linked. However, this is performed at another timing, for example, the activation of the application unit 21. It is also possible to switch the validity and invalidity of the task many times during the operation, and to control a plurality of tasks by one application.
[0219]
In this case, for a task that frequently switches between valid and invalid, if the memory is repeatedly reserved and released each time, the efficiency is low. Therefore, a flag (a flag indicating that the task is invalid) is set even after the invalidation. By itself, it is possible to keep the memory secured.
[0218]
Further, in the above-described example, it is assumed that nothing is stored in the common dictionary of the common dictionary unit 55 when the robot system is activated, but even if some words are stored in the common dictionary in advance. Good. For example, since the product name of the robot is often registered as the robot name, the product name of the robot may be registered in advance in the category of “_robot name_” in the common dictionary.
[0219]
FIG. 37 shows an example of a common dictionary when the robot product name "SDR" is entered in the category "_robot name_" when the robot system is activated. In FIG. 37, when the robot system is activated, the kana pronunciation “SDR” is entered in the category “_robot name_”. Therefore, the user can use the kana pronunciation “SDR” without registering the name. The robot can be controlled using the words represented.
[0220]
Also, in the above example, it was assumed that no cluster of unknown word acquisition units was generated at the initial stage (at the time of shipment). However, if a cluster is prepared for the main name from the beginning, when the name is input by voice in the name registration process of FIG. 21, it is easy to recognize the name for which the cluster is prepared. For example, if a cluster as shown in FIG. 3 is prepared at the time of shipment, the pronunciation of “red”, “blue”, “midori”, and “black” can be recognized (acquired) with a correct phoneme sequence. Further, it is not desirable that the pronunciation of a name for which a cluster is prepared is changed after the name is registered in the common dictionary. Therefore, when registering the pronunciation information in the common dictionary, instead of describing the cluster ID, the pronunciation is described in kana pronunciations (representative kana sequence of the cluster) such as “red”, “ao”, “midori”, and “kuro”. .
[0221]
Further, in the above-described example, the matching unit 54 reflects the contents of the common dictionary on all tasks. However, the matching unit 54 may reflect only the tasks to be reflected. For example, a number (task ID) is added to a task in advance, and the common dictionary in FIG. 24 is expanded to provide a column indicating “a list of tasks in which this entry is valid (or invalid)”, and the reflection in FIG. In the processing, the matching unit 54 only needs to reflect the contents of the common dictionary only on the task to which the task ID described in the column indicating “a list of valid tasks for this entry” is added.
[0222]
FIG. 38 shows an example in which the ID of the task to be reflected is described in a column indicating “valid tasks” in the common dictionary. In FIG. 38, the words belonging to the category “_robot name_” whose pronunciation in kana is represented by “SD” are valid task IDs “1”, “2”, and “4”. The contents of the common dictionary of words represented by the kana pronunciation “SD” are reflected only in the variable word dictionary 132 and the category table 133 of the tasks “1”, “2”, and “4”.
[0223]
Further, in the above-described example, the words stored in the fixed word dictionary 131 are words described in the language model 112, and the words stored in the variable word dictionary 132 are words belonging to the category. However, some of the words belonging to the category may be stored in the fixed word dictionary 131.
[0224]
FIG. 39 shows an application switching task 71. ₂ 40 shows an example of the category table 133 at the time of startup. That is, in the category table 133 in FIG. 40, the category “_robot name_” and the symbol “OOV00001” of the word belonging to the category “_robot name_” are registered in advance. Also, the fixed word dictionary 131 of FIG. 40 includes a symbol “OOV00001”, a transcription of a word represented by the symbol “OOV00001”, “ESD”, and a phoneme sequence “e / s / u / d / i”. : / A: / r / u "is registered in advance.
[0225]
In this case, the word “SDR” is subjected to the voice recognition process as belonging to the category “_robot name_”. That is, the word "SD" is treated as the robot name from the beginning. However, since the word "SD" is stored in the fixed word dictionary 131, it cannot be deleted or changed.
[0226]
In this way, for example, by storing words assumed to be set in a name, such as a product name of a robot, in the fixed word dictionary 131 in advance, the user can control the robot without performing name registration. Can be.
[0227]
In the above example, the category symbols are common to all tasks, but may not be common. In this case, a conversion table as shown in FIGS. 41 to 44 may be prepared in the task.
[0228]
That is, for example, when a certain task T describes a category “_ROBOT_NAME_” and a category “_USER_NAME_”, according to the conversion table in FIG. Are reflected in the category “_ROBOT_NAME_”. In the task T, the contents of the common dictionary of words belonging to the category “_user name_” are reflected in the category “_USER_NAME_”.
[0229]
For example, when a certain task T describes a category “_proper noun_”, according to the conversion table in FIG. 42, in the task T, the common dictionary unit 55 assigns the category “_robot name_”. Both the contents of the common dictionary of the words belonging thereto and the contents of the common dictionary of the words belonging to the category “_user name_” are reflected in the category “_proper noun_”.
[0230]
Further, for example, when a certain task T describes a category “_last name_” and a category “_first_”, according to the conversion table of FIG. The contents of the common dictionary of the words belonging to “_user name_” are converted and copied into a category “_surname_” and a category “_name_”. In the step of reflecting the contents of the common dictionary in this task (FIG. 25), for example, the second entry in FIG. 24 changes “_user name_cluster ID = 5” to “_last name_cluster ID = It is converted and duplicated into two entries of “5” and “_name_cluster ID = 5”, and then reflected in the fixed word dictionary and the category table of this task.
[0231]
Further, for example, when a category is not described in a certain task T, according to the conversion table in FIG. 44, in the task T, the category “_robot name_” and the category “_user name_” , The word belonging to the category “_place name_” is represented by the symbol “UNK”. Note that “UNK” means “Unknown word”.
[0232]
Thus, even in a task in which a category is not described, the matching unit 54 merely describes the symbol “UNK” in the language model 112 and the category “_robot name_” and the category “_user name_” , Words belonging to the category “_place name_” can be recognized.
[0233]
FIG. 45 shows an example of the external configuration of a bipedal walking robot provided with the robot control system 1 to which the present invention is applied. In the robot 201, a head unit 211 is provided above a body unit 213, and arm units 212 A and 212 B having the same configuration are provided on the left and right of the body unit 213, respectively. Leg units 214A and 214B having the same configuration are attached to predetermined positions on the lower left and right of 213, respectively.
[0234]
The head unit 211 includes CCD (Charge Coupled Device) cameras 221A and 221B functioning as “eyes” of the robot 201, microphones 222A and 222B functioning as “ears”, and a speaker 223 functioning as “mouth”. Are arranged at predetermined positions.
[0235]
FIG. 46 shows an example of the electrical configuration of the robot. The unit control system 231 and the dialogue control system 232 control the operation of the robot 201 according to a command from the robot control system 1. That is, the unit control system 231 controls the head unit 211, the arm units 212A and 212B, and the leg units 214A and 214B of the robot 201 as necessary, and causes the robot 201 to perform a predetermined operation. Further, the dialogue control system 232 controls the utterance of the robot 201 and causes the speaker 223 to make a predetermined utterance as necessary.
[0236]
In the above description, a word is a unit that should be treated as one unit in the process of recognizing speech, and does not always match a linguistic word. For example, “Taro-kun” may be treated as one word, or may be treated as two words “Taro” and “Kimi”. In addition, may be dealing with more is a major unit of "Hello Taro" or the like as one word.
[0237]
A phoneme is one that is more conveniently processed acoustically as one unit, and does not always match a phonetic phoneme or phoneme. For example, the "to" portion of "Tokyo" can be represented by three phonetic symbols "t / o / u", or a symbol "o:" which is a long sound of "o" is prepared. You may. Further, it may be expressed as “t / o / o”. In addition, symbols that represent silence are prepared, and they are further classified as "silence before utterance", "short silence section sandwiched between utterances", "silence in" tsu "part". Good.
[0238]
In the above description, the robot device has been described. However, the present invention can be applied to a device having an application using speech recognition, speech synthesis, translation, and other language processing.
[0239]
Further, the present invention can be applied to, for example, a device that extracts only predetermined terms from a dictionary registered in Kojien and creates the term dictionary.
[0240]
Further, in the above description, the case where there are a plurality of application units has been described, but one application unit may be provided.
[0241]
The above-described series of processes can be executed by hardware or can be executed by software. In this case, the above-described processing is executed by a personal computer 600 as shown in FIG.
[0242]
In FIG. 47, a CPU (Central Processing Unit) 601 performs various processes according to a program stored in a ROM (Read Only Memory) 602 or a program loaded from a storage unit 608 into a RAM (Random Access Memory) 603. Execute. The RAM 603 also appropriately stores data necessary for the CPU 601 to execute various processes.
[0243]
The CPU 601, the ROM 602, and the RAM 603 are mutually connected via an internal bus 604. The internal bus 604 is also connected to an input / output interface 605.
[0244]
The input / output interface 605 includes an input unit 606 including a keyboard and a mouse, a display including a CRT, an LCD (Liquid Crystal Display), an output unit 607 including a speaker, a storage unit 608 including a hard disk, and a modem. And a communication unit 609 including a terminal adapter and the like. The communication unit 609 performs communication processing via various networks including a telephone line and a CATV.
[0245]
A drive 610 is connected to the input / output interface 605 as necessary, and a removable medium 621 composed of a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is appropriately mounted. It is installed in the storage unit 608 as needed.
[0246]
When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer built in dedicated hardware or installing various programs. For example, it is installed on a general-purpose personal computer or the like from a network or a recording medium.
[0247]
As shown in FIG. 47, this recording medium is constituted not only by a package medium consisting of a removable medium 621 on which the program is recorded but also distributed to provide the user with the program, separately from the computer. And a hard disk including a ROM 602 and a storage unit 608 in which a program is recorded, which is provided to the user in a state where the program is incorporated in the apparatus main body in advance.
[0248]
In this specification, a step of describing a computer program refers to not only a process performed in chronological order according to the described order, but also a process executed in parallel or individually even if not necessarily performed in chronological order. Is also included.
[0249]
Also, in this specification, a system represents the entire device including a plurality of devices.
[0250]
Incidentally, Japanese Patent Application No. 2001-382579 previously proposed by the present applicant includes a series of processes until a word acquired by an unknown word acquiring mechanism is reflected on a language model, and a word reflected on a language model is referred to as a subsequent process. An invention is disclosed for a process for causing the recognition result to appear.
[0251]
However, the invention of Japanese Patent Application No. 2001-382579 is composed of an application for registering one word and an application using one registered word. It is assumed that there are a plurality of applications for performing voice recognition. In a system in which there are a plurality of applications and a variable number of applications, the above-described problem when reflecting registered words in the application and the above-described problem when deleting or changing words registered for a plurality of applications are described. It was difficult to solve the problem.
[0252]
Also, in Japanese Patent Application No. 2002-072718 previously proposed by the present applicant, the word “Taro”, which is an unknown word, is extracted from the utterance “My name is Taro (unknown word).” The invention of acquiring as a name is disclosed.
[0253]
However, the invention of Japanese Patent Application No. 2002-072718 discloses the above-described problem of reflecting an unknown word in a language model, the above-described problem of reflecting a registered word in an application, and a word registered in a plurality of applications. It has been difficult to solve the above-described problem when deleting or changing the.
[0254]
Therefore, in the invention of Japanese Patent Application No. 2001-382579 and the invention of Japanese Patent Application No. 2002-072718, the above-described problem of reflecting an unknown word in a language model, the above-described problem of reflecting a registered word in its application, In addition, it is difficult to solve all the problems described above when deleting or changing words registered for a plurality of applications.
[0255]
However, in the robot control system 1 of FIG. 1, since the category is described in the language model 112, it is possible to solve the above-described problem that the unknown word is reflected in the language model by making the unknown word belong to the category. Can be.
[0256]
Further, in the robot control system 1 of FIG. 1, the variable word dictionary 132 in which words to be subjected to speech recognition used in the application are registered or reconstructed based on the common dictionary. It is possible to solve the above-described problem when reflecting the word on the application and the problem when deleting or changing the word registered for a plurality of applications.
[0257]
Further, in the case of a system without a keyboard (for example, a robot or the like), there is a problem that it is difficult to input pronunciation information at the time of registering a word. As a means for solving the problem, for example, a phoneme typewriter is used. There has been proposed a method of inputting pronunciation information using voice.
[0258]
However, there is a problem that a phoneme typewriter may be erroneously recognized, and if the phoneme typewriter is used as it is, a word may be registered with an incorrect pronunciation. For example, if the pronunciation of "ESD" is recognized by the phoneme typewriter and the phoneme typewriter misrecognizes and outputs a result of "Irnyal", if "Irnyal" is adopted as the pronunciation information, an incorrect pronunciation will be made. in order to word is registered, and later, for example, "es Dr, Hello" that although the speech is difficult to be recognized, "Isuniyaru, Hello" speech that the context of recognized easily occurs.
[0259]
Therefore, in the robot control system 1 of FIG. 1, the latest pronunciation information at that time is acquired from the unknown word acquisition unit 56 every time a word registered by voice is reflected on each task, so that the phoneme typewriter is erroneous. Even after the recognized and misrecognized word is registered in the variable word dictionary 132, the pronunciation information is updated only by supplying the speech data to the unknown word acquisition unit 56, and the latest pronunciation information at that time can be obtained. , Normal recognition results may be obtained.
[0260]
That is, in the robot control system 1 of FIG. 1, the above-described problem of reflecting an unknown word in a language model, the above-described problem of reflecting a registered word in an application, and a word registered in a plurality of applications are referred to. The above-described problem when deleting or changing, and the above-mentioned problem when inputting pronunciation information of a word to be registered in a system without a keyboard, that is, all the above-described problems can be solved.
[0261]
【The invention's effect】
As described above, according to the present invention, words can be registered. In particular, even when words corresponding to a plurality of applications are registered, the registered words can be commonly used in each application. Also, words registered before the application is started can be used in the application. Further, even when the registered word is changed, consistency can be maintained in each application.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration example of a robot control system to which the present invention has been applied.
FIG. 2 is a block diagram illustrating a configuration example of a speech recognition engine unit in FIG. 1;
FIG. 3 is a diagram illustrating an example of a cluster of an unknown word acquisition unit in FIG. 2;
FIG. 4 is a flowchart illustrating a robot control process in the robot control system of FIG. 1;
FIG. 5 is a flowchart illustrating a robot control process in the robot control system of FIG. 1;
FIG. 6 is a diagram illustrating a configuration example of a task in FIG. 2;
FIG. 7 is a diagram showing an example of a phoneme list in FIG. 6;
FIG. 8 is a diagram showing an example of kana phoneme conversion rules of FIG. 6;
FIG. 9 is a diagram showing an example of a language model of the phoneme typewriter task of FIG. 2;
FIG. 10 is a diagram showing an example of a fixed word dictionary of the phoneme typewriter task of FIG. 2;
11 is a diagram illustrating an example of a language model of the application switching task in FIG. 2;
FIG. 12 is a diagram illustrating an example of a fixed word dictionary of the application switching task in FIG. 2;
FIG. 13 is a diagram illustrating an example of a category table of the application switching task in FIG. 2;
FIG. 14 is a diagram illustrating an example of a language model of the name registration task in FIG. 2;
FIG. 15 is a diagram illustrating an example of a fixed word dictionary of the name registration task in FIG. 2;
FIG. 16 is a diagram illustrating an example of a language model of the chat task of FIG. 2;
17 is a diagram illustrating an example of a fixed word dictionary of the chat task of FIG. 2;
FIG. 18 is a diagram illustrating an example of a category table of the chat task of FIG. 2;
19 is a diagram illustrating an example of a language model of the voice commander task in FIG. 2;
20 is a diagram showing an example of a fixed word dictionary of the voice commander task of FIG. 2;
FIG. 21 is a flowchart illustrating a name registration process in step S9 of FIG. 5;
FIG. 22 is a flowchart illustrating a name recognition process in step S43 of FIG. 21;
FIG. 23 is a diagram illustrating an example of a cluster of an unknown word acquisition unit in FIG. 2;
FIG. 24 is a diagram illustrating an example of a common dictionary unit in FIG. 2;
FIG. 25 is a flowchart illustrating a reflection process in step S49 of FIG. 21;
FIG. 26 is a block diagram illustrating a reflection process of FIG. 25;
FIG. 27 is a diagram showing an example of a variable word dictionary of the name registration task of FIG. 2;
FIG. 28 is a diagram illustrating an example of a category table of the name registration task in FIG. 2;
FIG. 29 is a flowchart illustrating a process of deleting or changing a word in the matching unit of FIG. 2;
FIG. 30 is a diagram showing an example of changing the common dictionary unit of FIG. 2;
FIG. 31 is a diagram illustrating an example of changing the common dictionary unit in FIG. 2;
FIG. 32 is a flowchart illustrating the chat processing of step S12 in FIG. 5;
FIG. 33 is an example of a variable word dictionary of the chat task of FIG. 2;
FIG. 34 is an example of a category table of the chat task in FIG. 2;
FIG. 35 is a flowchart illustrating a speech recognition process in step S123 of FIG. 32;
FIG. 36 is a diagram illustrating an example of a formula for calculating a language score.
FIG. 37 is a diagram illustrating a modification of the common dictionary unit in FIG. 2;
FIG. 38 is a diagram illustrating a modification of the common dictionary unit in FIG. 2;
FIG. 39 is a diagram showing a modification of the fixed word dictionary of FIG. 6;
FIG. 40 is a diagram showing an example of the category table of FIG. 6;
FIG. 41 is a diagram showing an example of a category conversion table.
FIG. 42 is a diagram illustrating an example of a category conversion table.
FIG. 43 is a diagram showing an example of a category conversion table.
FIG. 44 is a diagram showing an example of a category conversion table.
FIG. 45 is a perspective view showing an external configuration of the robot.
FIG. 46 is a block diagram illustrating an electrical configuration of the robot.
FIG. 47 is a diagram illustrating an example of a personal computer.
[Explanation of symbols]
11 Speech Recognition Engine Unit, 21 Application Unit, 31 Application Management Unit, 51 Microphone, 52 AD Conversion Unit, 53 Feature Extraction Unit, 54 Matching Unit, 55 Common Dictionary Unit, 56 Unknown Word Acquisition Unit, 71 Task, 111 Acoustic Model , 112 language models, 113 dictionaries, 114 phoneme lists, 115 kana phoneme conversion rules, 116 search parameters, 131 fixed word dictionaries, 132 variable word dictionaries, 133 category tables

Claims (8)

A language processing apparatus having an application using language processing,
A registered dictionary storage means for storing a registered dictionary in which words are registered;
A language processing apparatus comprising: a construction unit configured to construct, based on the registered dictionary, a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered. Processing means for adding, deleting, or changing a word to the registered dictionary;
A deleting unit for deleting a word in the dedicated dictionary,
After all words registered in the dedicated dictionary have been deleted,
2. The language processing apparatus according to claim 1, wherein the construction unit reconstructs the dedicated dictionary based on the registered dictionary in which a word has been added, deleted, or changed. The dedicated dictionary includes at least a fixed dictionary in which predetermined words are registered in advance, and a variable dictionary in which words to be registered are variable,
2. The language processing apparatus according to claim 1, wherein the construction unit constructs the variable dictionary among the dedicated dictionaries. The dedicated dictionary further includes a category table in which categories of words are registered,
4. The variable dictionary according to claim 3, wherein the constructing unit constructs the variable dictionary by registering, in the variable dictionary, words of a category registered in the category table among words of the registered dictionary. Language processor. There are a plurality of the applications,
The language processing apparatus according to claim 1, wherein the construction unit constructs the dedicated dictionary for each of the plurality of applications. A language processing method of a language processing apparatus having an application that uses language processing,
A registered dictionary storage step for storing a registered dictionary in which words are registered;
A construction step of constructing, based on the registered dictionary, a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered. A language processing program for an application,
A computer is characterized by including a construction step of constructing a dedicated dictionary dedicated to the application in which words to be subjected to language processing used in the application are registered based on the registered dictionary in which words are registered. A program storage medium that stores a readable program. A language processing program for an application,
A word to be subjected to linguistic processing used in the application is registered. The computer executes a construction step of constructing a dedicated dictionary dedicated to the application based on the registered dictionary in which words are registered. Program to do.

Images