JP2003271180A - Voice processor, voice processing method, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enable a user to register an unknown word without caring a registration mode during continuous normal conversation in an interactive system. <P>SOLUTION: A feature value extracting part 43 extracts the feature parameter of a voice input fetched from a microphone 41 through an AD converting part 42. A phoneme typewriter 45 outputs a phoneme sequence based on the feature parameter. A matching part 44 refers to a dictionary database 52 to generate a word string based on the feature parameter and estimates an unknown word. A conversation control part 3 registers a combination of the unknown word included in the word string and a category estimated from the word string in an association storing part 2 when the word string corresponds to a prescribed template. This present invention can be applied to an interactive system that performs conversation based on voice recognition processing. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice processing device, a voice processing method, a program, and a recording medium, and more particularly, to an input voice signal during voice recognition of continuously input voice signals. The present invention relates to a voice processing device and a voice processing method, a program, and a recording medium that can extract an unknown word included in and easily register it.

[0002]

2. Description of the Related Art In interactive systems, there are many situations in which a name is registered by voice. For example, a user may register his / her name, name the dialog system, or enter a place name or a shop name.

Conventionally, as a method of easily realizing such voice registration, there is a method of shifting to the registration mode by some command and returning to the normal interactive mode when the registration is completed. In this case, for example, "User name registration"
When the user generates a name, the voice command is entered to register the mode, and thereafter, the name is registered, and then the process returns to the normal mode.

[0004]

However, in such a voice registration method, there is a problem in that the mode must be switched by a command, which is unnatural as a dialogue and annoying to the user. Further, when there are a plurality of objects to be named, the number of commands increases, which makes it even more troublesome.

[0005] Further, in the registration mode, if the user is you've talked a word other than the name (for example, "Hello"), it will also be registered as a name. Also, for example, when the user speaks with a word other than the name, such as "My name is Taro." Instead of just the name "Taro," the whole ("My name is Taro. )) Is registered as a name.

The present invention has been made in view of such a situation, and an object thereof is to enable a user to register a word in a normal dialogue without making the user aware of the registration mode.

[0007]

A speech processing apparatus of the present invention determines a recognition means for recognizing continuous input speech and whether or not an unknown word is included in the recognition result recognized by the recognition means. When the unknown word determination means and the unknown word determination means determine that the unknown word is included, the acquisition means for acquiring the word corresponding to the unknown word, and the word acquired by the acquisition means as other information. It is characterized by comprising a registration means for registering in association with each other.

When the recognition result is judged to match the specific pattern by the pattern judging means, the registration means further comprises pattern judging means for judging whether or not the recognition result matches the specific pattern. , You can register words.

If the unknown word determination means determines that the unknown word is not included, or if the pattern determination means determines that the recognition result does not match a specific pattern, it corresponds to the input voice. It may further comprise a response generating means for generating a response to

The registration means can register a word category as other information.

The registration means can register other information based on the pattern judged to match by the pattern judgment means.

The acquisition means can acquire words by clustering unknown words.

[0013] For a predetermined section of the input voice, there is further provided comparison means for comparing acoustic scores in the case of matching with a known word and in the case of being recognized by the phoneme typewriter, and the comparing means is made to recognize by the phoneme typewriter. If the acoustic score is better, the section can be estimated to be an unknown word.

The comparison means can make a comparison after correcting the acoustic score when the phonological typewriter recognizes the acoustic score when the known word is matched.

The speech processing method of the present invention comprises a recognition step of recognizing continuous input speech, a determination step of determining whether or not an unknown word is included in the recognition result recognized by the processing of the recognition step, When it is determined by the processing of the determination step that the unknown word is included, the acquisition step of acquiring the word corresponding to the unknown word and the word acquired by the processing of the acquisition step are registered in association with other information. And a registration step.

The program of the recording medium of the present invention comprises a recognition step of recognizing continuous input speech, and a judgment step of judging whether or not the recognition result recognized by the processing of the recognition step includes an unknown word. If the process of the determination step determines that the unknown word is included, the acquisition step of acquiring the word corresponding to the unknown word and the word obtained by the process of the acquisition step are registered in association with other information. And a registration step to perform.

The program of the present invention comprises a recognition step of recognizing continuous input speech, a judgment step of judging whether or not an unknown word is included in the recognition result recognized by the processing of the recognition step, and a judgment step. By the processing of
When it is determined that the unknown word is included, the acquisition step of acquiring the word corresponding to the unknown word and the registration step of registering the word acquired by the processing of the acquisition step in association with other information are executed. It is characterized by

In the present invention, when continuous input speech is recognized and the recognition result includes an unknown word, a word corresponding to the unknown word is acquired and the word is registered in association with other information. It

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a configuration example of an embodiment of a dialogue system to which the present invention is applied.

This dialogue system is a system for dialogue with a user (human) by voice. For example, when voice is input, a name is extracted from the voice and registered.

That is, the voice recognition unit 1 is adapted to receive a voice signal based on the utterance from the user, and the voice recognition unit 1 recognizes the inputted voice signal and outputs the result of the voice recognition. Text as, and other accompanying information,
Output to the dialogue control unit 3 and the word acquisition unit 4 as necessary.

The word acquisition unit 4 automatically stores the acoustic characteristics of a word that is not registered in the recognition dictionary of the voice recognition unit 1 so that the voice of the word can be recognized thereafter.

That is, the word acquisition unit 4 obtains the pronunciation corresponding to the input voice by the phoneme typewriter and classifies it into several clusters. Each cluster has an ID and a representative phoneme sequence and is managed by the ID. The state of the cluster at this time will be described with reference to FIG.

For example, "red", "blue", "green"
It is assumed that there are three input voices. In this case, the word acquisition unit 4 outputs three times of voices to the “red” cluster 21, the “blue” cluster 22, and the “green” corresponding to each of the three sounds.
The cluster 23 is classified into three clusters, and each cluster has a representative phoneme sequence (in the example of FIG. 2, "a / k / a,
“A / o”, “m / i / d / o / r / I”) and ID (in the case of FIG. 2,
"1", "2", "3") are added.

When the voice "red" is input again, since the corresponding cluster already exists, the word acquisition unit 4 classifies the input voice into the "red" cluster 21,
Does not create a new cluster. On the other hand, "black"
When the voice is input, since the corresponding cluster does not exist, the word acquisition unit 4 newly generates a cluster 24 corresponding to “black”, and a representative phoneme sequence (in FIG. 2) is generated in the cluster. In the case of the example, "k / u / r / o") and ID (Fig. 2)
In the case of the above example, “4”) is added.

Therefore, whether or not the input voice is an unacquired word can be determined by whether or not a new cluster is generated. The details of such word acquisition processing are described in Japanese Patent Application No. 2001-97843 previously proposed by the present applicant.
No.

The associative storage unit 2 stores the registered name (unknown word).
It stores information such as a category such as whether it is a user name or a character name. For example, in the example of FIG. 3, the cluster ID and the category name are stored in association with each other.
In the case of the example in FIG. 3, for example, the cluster ID “1”,
"3" and "4" correspond to the "user name" category,
The cluster ID “2” corresponds to the “character name” category.

The dialogue control unit 3 understands the content of the user's utterance from the output of the voice recognition unit 1 and controls the registration of the name (unknown word) based on the result of the understanding. Further, the dialogue control unit 3 controls the subsequent dialogue so that the registered name can be recognized based on the registered name information stored in the associative storage unit 2.

FIG. 4 shows a configuration example of the voice recognition unit 1.

The utterance of the user is input to the microphone 41, and the microphone 41 converts the utterance into a voice signal as an electric signal. This audio signal is AD
It is supplied to the (Analog Digital) converter 42. The AD converter 42 samples an audio signal which is an analog signal from the microphone 41, quantizes it, and converts it into audio data which is a digital signal. This voice data is supplied to the feature amount extraction unit 43.

The feature amount extraction unit 43 extracts feature parameters such as a spectrum, a power linear prediction coefficient, a cepstrum coefficient, and a line spectrum pair for each appropriate frame from the voice data from the AD conversion unit 42, and a matching unit. 44 and the phoneme typewriter unit 45.

The matching unit 44 refers to the acoustic model database 51, the dictionary database 52, and the language model database 53 as needed based on the characteristic parameters from the characteristic amount extraction unit 43, and inputs the voice input to the microphone 41. Find the word string closest to (input voice).

The acoustic model database 51 stores acoustic models representing acoustic features such as individual phonemes and syllables in the language of speech to be recognized. As the acoustic model, for example, an HMM (Hidden Markov Model) or the like can be used. The dictionary database 52 stores, for each word (phrase) to be recognized, a word dictionary in which information regarding pronunciation is described, and a model in which a phoneme or a syllable chain relation is described.

The word here is a unit that is more convenient to handle as a unit in the recognition process, and does not necessarily match a linguistic word. For example, "Taro-kun" may be treated as one word as a whole, 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.

The phoneme means that it is more convenient to process acoustically as one unit in terms of processing, and does not necessarily correspond to a phonetic phoneme or phoneme. For example,
The "to" part of "Tokyo" may be represented by the three phoneme symbols "t / o / u", or "t / o" using the symbol "o:" which is the long consonant of "o". It may be expressed as ":". Alternatively, it can be expressed as "t / o / o". In addition, we also prepared a symbol that represents silence, and also used it as "silence before utterance", "short silence interval between utterances", "silence of utterances", and "silence of" tsu "part". It may be finely categorized as shown in FIG.

The language model database 53 describes information about how each word registered in the word dictionary of the dictionary database 52 is chained (connected).

The phoneme typewriter unit 45 acquires a phoneme sequence corresponding to the inputted voice, based on the characteristic parameters supplied from the characteristic amount extraction unit 43. For example, from the voice saying "My name is Taro.""W / a / t / a / sh / i / n /
o / n / a / m / a / e / w / a / t / a / r / o: / d / e / s / u "is obtained. This phoneme typewriter has the existing ones. Can be used.

Note that a phoneme typewriter other than the phoneme typewriter can be used instead as long as the phoneme sequence can be obtained for any voice. For example, speech recognition using Japanese syllables (a, i, u ... ka, ki ... n) as a unit, or speech recognition using subwords that are larger than phonemes and smaller than words as units. It is also possible.

The control unit 46 includes an AD conversion unit 42, a feature amount extraction unit 43, a matching unit 44, and a phoneme typewriter unit 45.
Control the behavior of.

Next, referring to the flow chart of FIG.
The processing of the dialogue system of the present invention will be described.

In step S21, when the user inputs a voice into the microphone 41, the microphone 41
The utterance is converted into a voice signal as an electric signal. Then, in step S22, the voice recognition unit 1 executes a voice recognition process.

Details of the voice recognition process will be described with reference to FIG. The audio signal generated by the microphone 41 is processed by the AD conversion unit 42 in step S41.
It is converted into voice data which is a digital signal and supplied to the feature amount extraction unit 43.

In step S42, the feature quantity extraction unit 4
3 receives the audio data from the AD converter 42. Then, the feature amount extraction unit 43 proceeds to step S43, and extracts the feature parameters such as the spectrum, the power, and the time change amount thereof for each appropriate frame, and supplies the feature parameters to the matching unit 44.

In step S44, the matching unit 4
4 connects some of the word models stored in the dictionary database 52 to generate a word string. In addition, not only known words registered in the dictionary database 52 but also "<OOV>", which is a symbol representing an unregistered unknown word, is included in the words forming the word string. This word string generation process will be described in detail with reference to FIG.

In step S61, the matching unit 4
4 calculates the acoustic score in both cases for a certain section of the input voice. That is, the acoustic score obtained as a result of matching with the known word registered in the dictionary database 52 and the result obtained by the phonological typewriter unit 45 (in this case, "w / a / t / a / sh / i / n / o / n / a / m / a / e / w / a / t / a / r / o: / d / e
The acoustic score of each part of "/ s / u") is calculated. The acoustic score represents how close the word string that is the candidate of the speech recognition result and the input speech are as sounds.

Next, the acoustic score obtained as a result of matching a partial section of the input voice with the known word registered in the dictionary database 52 is compared with the acoustic score obtained as a result by the phonological typewriter unit 45. , Matching with known words is performed on a word-by-word basis, and the phoneme typewriter unit 45
Since matching is performed in phoneme units and the scales are different, it is difficult to compare as it is (generally, the acoustic score in phoneme unit has a larger value). Therefore, in order to make it possible to compare the scales, the matching unit 44, in step S62,
The acoustic score of the result obtained by the phonological typewriter unit 45 is corrected.

For example, a process of multiplying the acoustic score from the phoneme typewriter unit 45 by a coefficient or subtracting a constant value or a value proportional to the frame length is performed. Of course, since this process is relative, it can be performed on the acoustic score resulting from matching with a known word. In addition,
For details of this process, refer to the document "" EUROSPEECH99 Volu
"Me 1, Page 49-52" to "OOV-Detection in Large Voca
bulary System Using Automatically Defined Word-Frag
ments as Fillers ”.

In step S63, the matching section 44 compares the two acoustic scores (determines whether the acoustic score obtained by the phonological typewriter section 45 is higher (excellent)). . When the acoustic score of the result recognized by the phonological typewriter unit 45 is higher,
In step S64, the matching unit 44 estimates that the section is <OOV> (Out Of Vocabulary) (unknown word).

If it is determined in step S63 that the acoustic score of the result recognized by the phonological typewriter unit 45 is lower than the acoustic score of the result of matching the known word, the process proceeds to step S66. Part 4
4 estimates that the section is a known word.

That is, for example, for the section corresponding to "Taro", "t / a / r /" output from the phoneme typewriter unit 45.
Compare the acoustic score of "o:" with the acoustic score when matching with a known word, and if the acoustic score of "t / a / r / o:" is higher, the word corresponding to that voice section As "<O
"OV> (t / a / r / o :)" is output, and when the acoustic score of the known word is higher, the known word is output as a word corresponding to the voice section.

In step S65, n word strings (concatenation of several word models) which are estimated to have a high acoustic score are preferentially generated.

Returning to FIG. 6, in step S45,
The phoneme typewriter unit 45 performs the recognition in units of phonemes on the feature parameters extracted in the process of step S43 independently of the process of step S44, and outputs a phoneme sequence. For example, when the voice "My name is Taro (unknown word)" is input, the phonological typewriter unit 45 displays "w / a / t / a / sh / i / n / o / n / a". / m / a / e / w / a / t / a / r / o: / d
The phoneme sequence "/ e / s / u" is output.

In step S46, the matching unit 4
4 calculates the acoustic score for each word string generated in step S44. For word strings that do not contain <OOV> (unknown words), the existing method is used, that is, the likelihood is calculated by inputting the speech feature parameters for each word string (concatenation of word models). Use the method. On the other hand, <O
For word strings containing OV>, it is not possible to obtain the acoustic score of the voice section corresponding to <OOV> by the existing method (since there is no word model corresponding to <OOV> in advance). Therefore, for that voice section, the acoustic score of the same section is extracted from the recognition result of the phonological typewriter, and the corrected value is adopted as the <OOV> acoustic score. Further, it is integrated with the acoustic scores of other known word parts and used as the acoustic score of the word string.

In step S47, the matching unit 4
In No. 4, the upper m (m ≦ n) word strings having a high acoustic score are left as candidate word strings. In step S48, the matching unit 44 refers to the language model database 53 and calculates a language score for each candidate word string. The language score represents how appropriate a word string that is a candidate of the recognition result is as a word. Here, the method of calculating the language score will be described in detail.

Since the speech recognition unit 1 of the present invention recognizes unknown words as well, the language model needs to correspond to the unknown words.
For example, using a grammar or Finite State Automaton (FSA) corresponding to an unknown word and using a tri-gram (one of statistical language models) corresponding to an unknown word. Will be described.

An example of the grammar will be described with reference to FIG. This grammar 61 is described in BNF (Backus Naur Form).
In FIG. 9, "$ A" represents "variable" and "A | B" represents "A or B". Also, "[A]" means "A can be omitted", and {A} is "A is 0".
Means "to repeat more than once".

<OOV> is a symbol representing an unknown word. By describing <OOV> in the grammar, it is possible to deal with a word string including an unknown word. "$ ACTION" is Figure 9
However, in reality, names of actions such as “standing”, “seating”, “bowing”, and “greeting” are defined.

[0058] In the grammar 61, "<start> / Hello / <end>" ( "/" is a separator between words), "<head>
/ Goodbye / <End> ”,“ <Top> / I / Name / wa / <
Word strings that match the grammar stored in the database such as "OOV> / is / <end>" are accepted (parsed by this grammar), but "<head> / Kimi / no / <OOV>" / Name / <
Term strings that do not match the grammar stored in the database, such as ">", are not accepted (not parsed by this grammar). It should be noted that "<start>" and "<end>" are special symbols representing silence before and after utterance, respectively.

A parser (analyzer) is used to calculate a language score using this grammar. The parser divides the word string into word strings that can accept the grammar and word strings that cannot accept the grammar. That is, for example, a linguistic score of 1 is given to an acceptable word string, and a linguistic score of 0 is given to an unacceptable word string.

Therefore, for example, "<start> / me / no /
Name / is / <OOV> (t / a / r / o:) / is / <Termination> ”and“ <
If there are two word strings "start> / me / of / name / is / <OOV> (j / i / r / o:) // <end>", both are "<
The top> / me / of / name / is / <OOV> // <end> ”and the language score is calculated, and the language score 1 (accepted) is output together.

To determine whether or not the grammar of the word string can be accepted, the grammar is converted in advance to an equivalent (or approximate) finite state automaton (hereinafter referred to as FSA),
It can also be realized by determining whether each word string can be accepted by the FSA.

An example of converting the grammar of FIG. 9 into an equivalent FSA is
It is shown in FIG. FSA is a directed graph consisting of states (nodes) and paths (arcs). As shown in FIG. 10, S1 is the start state and S16 is the end state. Further, the name of the action is actually registered in "$ ACTION", as in FIG.

A word is attached to the path, and the path consumes this word when the state transits from a predetermined state to the next state. However, a path to which "ε" is added is a special transition that does not consume words (hereinafter referred to as "ε transition").
That is, for example, in “<head> / I / was / <OOV> / is / <end>”, the initial state S1 transits to state S2, the <head> is consumed, and state S2 changes to state S3. A transition occurs and “I” is consumed, but since the transition from the state S3 to the state S5 is an ε transition, no word is consumed. That is,
It is possible to skip from the state S3 to the state S5 and transit to the next state S6.

Whether or not a predetermined word string can be accepted by this FSA is determined by starting from the initial state S1 and reaching the end state S16.

That is, for example, "<start> / I / no / name / is
In "/ <OOV> / is / <end>", the word "<start>" is consumed by making a transition from the initial state S1 to the state S2.
Next, the state "S2" transits to the state S3, and the word "I" is consumed. Similarly, from state S3 to state S4,
The state S4 transits to the state S5, the state S5 transits to the state S6, and the state S6 transits to the state S7, and "no", "name", "ha", and "<00V>" are successively consumed. Further, the state S7 transits to the state S15, "is" is consumed, the state S15 transits to the state S16, "<end>" is consumed, and the final state S16 is eventually reached.
Therefore, "<start> / me / of / name / is / <OOV> / is / <
Termination> ”is accepted by the FSA.

However, "<top> / Kimi / no / <OOV> /
Name / <end> ”is changed from state S1 to state S2, state S
From 2 to state S8, and from state S8 to state S9, "<head>", "kun", and "no" are consumed, but transition to the end state S16 is possible because the transition cannot be made beyond that. You cannot do it. Therefore, "<top> / Kimi / of
/ <OOV> / Name / <Termination> ”is not accepted by the FSA (not accepted).

Further, an example of calculating a language score when a tri-gram which is one of statistical language models is used as a language model will be described with reference to FIG. The statistical language model is a language model in which the generation probability of the word string is obtained and used as the language score. That is, for example, the language score of “<first> / me / of / name / is / <OOV> // <end>” of the language model 71 of FIG. 11 is the word as shown in the second line. It is represented by the probability of column generation. This is
It is expressed as the product of conditional probabilities, as shown in lines 3-6. Note that, for example, “P (| <start>”
"I)" means that the word immediately before "no" is "I" and the word immediately before "I" is "<start>",
It represents the probability that "no" will appear.

Further, in the tri-gram, the equations shown in the third to sixth lines of FIG. 11 are approximated by the conditional probability of three consecutive words, as shown in the seventh to ninth lines. .
These probability values are obtained by referring to the tri-gram database 81 as shown in FIG. The tri-gram database 81 is obtained by analyzing a large amount of texts in advance.

In the example of FIG. 12, three consecutive words w
The probability P (w3 | w1w2) of 1, w2, w3 is represented. For example, if the three words w1, w2, and w3 are "<start>", "I", and "no", the probability value is 0.12, and "me", "no", and "name" are set. , The probability value is 0.01, “<OOV>”, “is”,
In the case of “<end>”, the probability value is set to 0.87.

Of course, "P (W)" and "P (w2 | w
Similarly, “1)” is also obtained in advance.

In this way, by performing entry processing for <OOV> in the language model, the language score can be calculated for the word string including <OOV>.
Therefore, the symbol <OOV> can be output as the recognition result.

Also, in the case of using another type of language model, the language score can be calculated for the word string including <OOV> similarly by performing the entry process for <OOV>.

Furthermore, even when a language model in which an <OOV> entry does not exist is used, the language score can be calculated by using a mechanism that maps <OOV> to an appropriate word in the language model. For example, "P (<OOV> | I am)"
Even if you use a tri-gram database that does not exist,
Access the database with "P (Taro ｜ I am)" and set the probability described in it to "P (<OOV> ｜ I
The language score can be calculated by regarding the value as "ha)".

Returning to FIG. 6, the matching unit 44 integrates the acoustic score and the language score in step S49. In step S50, the matching unit 44 selects a candidate word string having the best score based on the score obtained by integrating both the acoustic score and the language score obtained in step S49, and outputs the candidate word string as a recognition result.

When the finite state automaton is used as the language model, the integration process of step S49 is executed. When the language score is 0, the word string is deleted, and when the language score is other than 0, it is left as it is. May be processed.

Returning to FIG. 5, after the voice recognition processing is executed in step S22 as described above, step S23 is performed.
At, the control unit 46 of the voice recognition unit 1 determines whether the recognized word string includes an unknown word. When it is determined that the unknown word is included, the control unit 46 controls the word acquisition unit 4 to execute the word acquisition process in step S24 to acquire the unknown word.

Details of the word acquisition process will be described with reference to FIG. In step S91, the word acquisition unit 4 extracts the characteristic parameter of the unknown word (<OOV>) from the voice recognition unit 1. In step S92, the word acquisition unit 4 determines whether the unknown word belongs to the already acquired cluster. If it is judged that it does not belong to the already acquired cluster,
The word acquisition part 4 produces | generates the new cluster corresponding to the unknown word in step S93. Then, in step S94, the word acquisition unit 4 outputs the ID of the cluster to which the unknown word belongs to the matching unit 44 of the voice recognition unit 1.

If it is determined in step S92 that the unknown word belongs to the already-acquired cluster, it is not necessary to generate a new cluster.
The process of step 93 is skipped, and the process proceeds to step S94, where the matching unit 4 determines the ID of the acquired cluster to which the unknown word belongs.
Output to 4.

The process of FIG. 13 is performed for each unknown word.

Returning to FIG. 5, after the word acquisition processing in step S24 is completed, in step S25, the dialogue control unit 3
Determines whether the word string acquired in the process of step S24 matches the template. That is,
Here, it is determined whether or not the word string of the recognition result means registration of some name. And step S2
When it is determined in step 5 that the word string of the recognition result matches the template, in step S26,
The dialogue control unit 3 stores the cluster ID of the name in the associative storage unit 2.
And the categories are stored in association with each other.

An example of a template matched by the dialogue control unit 3 will be described with reference to FIG. In FIG. 14, “/ A /” means “if the character string A is included”, and “A | B” means “A or B”. Also, "." Represents "arbitrary character" and "A +" is "A
Is repeated one or more times, and "(.) +" Represents "arbitrary character string".

This template 91 represents that the operation on the right side of the figure is executed when the word string of the recognition result matches the regular expression on the left side of the figure. For example, the recognition result is "<first> / me / of / name / is / <OOV> (t / a / r / o:) /" /
In the case of the word string “<end>”, the character string “My name is <OOV>” generated from this recognition result is shown in FIG.
Matches the second regular expression of. Therefore, the corresponding operation of “registering the cluster ID corresponding to <OOV> as a user name” is executed. That is, "<OOV
When the cluster ID of "> (t / a / r / o :)" is "1", the category name of the cluster ID "1" is registered as the "user name" as shown in FIG.

Further, for example, the recognition result is "<head> /
If you / 's / name / is / <OOV> (a / i / b / o) /' / <end>', the string "Your name is <OOV>"
14 matches the first regular expression in FIG. 14, so if “<OOV> (a / i / b / o)” is the cluster ID “2”, the category of the cluster ID “2” is , "Character name" is registered.

Depending on the interactive system, there may be only one type of word to be registered (for example, only “user name”). In that case, the template 91 and the associative storage unit 2 can be simplified. For example, template 9
It is possible to store only the cluster ID in the associative storage unit 2 by setting the content of 1 to “store the ID if <OOV> is included in the recognition result”.

The dialogue control unit 3 reflects the information thus registered in the associative storage unit 2 in the subsequent dialogue judgment processing. For example, on the side of the dialogue system, "determine whether or not the name of the dialogue character is included in the user's utterance. Process "
When the process of “the dialogue character speaks the user's name” becomes necessary, the dialogue control unit 3 causes the associative storage unit 2
By referring to the information of (1), it is possible to obtain the word corresponding to the dialogue character (entry whose category name is "character name") or the word corresponding to the user name (entry whose category name is "user name"). .

On the other hand, if it is determined in step S23 that the recognition result does not include an unknown word, or if it is determined in step S25 that the recognition result does not match the template, the dialogue is executed in step S27. The control unit 3 generates a response corresponding to the input voice. That is, in this case, the name (unknown word) registration process is not performed, and the predetermined process corresponding to the input voice from the user is performed.

By the way, when a grammar is used as a language model, a description corresponding to a phoneme typewriter can be incorporated in the grammar. An example of the grammar in this case is shown in FIG. In this grammar 101, the variable "$ PH" on the first line
"ONEME" means any one of the phonological symbols because all phonemes are connected by "|" which means "or". The variable "OOV" indicates that "$ PHONEME" is repeated 0 times or more. That is, it means "a phoneme symbol connected zero or more times" and corresponds to a phoneme typewriter. Therefore, "$ OOV" between "ha" and "is" in the 3rd line is
Any pronunciation can be accepted.

In the recognition result when this grammar 101 is used, the portion corresponding to "$ OOV" is output as a plurality of symbols. For example, the recognition result of "My name is Taro" is "<
Start> / Me // Name / is / t / a / r / o: // <End> ”. The result is "<top> / I / name / was / <OOV> (t / a /
r / o:) / is ”, the processes after step S23 in FIG. 5 can be executed in the same manner as in the case of using the phoneme typewriter.

In the above description, the category is registered as the information related to the unknown word, but other information may be registered.

FIG. 16 shows an example of the configuration of the personal computer 110 that executes the above processing. This personal computer 110 has a CPU (Central Processin)
g Unit) 111 is built in. Bus 1 for CPU111
An input / output interface 115 is connected via 14. A ROM (Read Only Memory) 11 is provided on the bus 114.
2 and a RAM (Random Access Memory) 113 are connected.

The input / output interface 115 includes a mouse operated by the user, a keyboard, a microphone, and an AD.
An input unit 117 including an input device such as a converter and an output unit 116 including an output device such as a display, a speaker, and a DA converter are connected. Further, the input / output interface 115 is connected to a storage unit 118 including a hard disk drive for storing programs and various data, and a communication unit 119 for communicating data via a network typified by the Internet.

The input / output interface 115 includes a magnetic disk 131, an optical disk 132, and a magneto-optical disk 1.
A drive 120 for reading / writing data from / to a recording medium such as 33 and a semiconductor memory 134 is connected as necessary.

An audio processing program for causing the personal computer 110 to execute the operation as an audio processing apparatus to which the present invention is applied is a magnetic disk 131 (including a floppy disk), an optical disk 132 (CD-ROM (Compact Disk
sc-Read Only Memory), DVD (Digital Versatile Disc)
Drive), the magneto-optical disc 133 (including MD (Mini Disc)), or the semiconductor memory 134, and is supplied to the personal computer 110 to drive the drive 1
It is read by 20 and installed in the hard disk drive built in the storage unit 118. The voice processing program installed in the storage unit 118 is transferred from the storage unit 118 to the RAM 11 according to a command from the CPU 111 corresponding to a command input by the user to the input unit 117.
3 is loaded and executed.

The series of processes described above can be executed by hardware, but can also be executed by software. When a series of processes is executed by software, a program that constitutes the software executes a variety of functions by installing a computer in which dedicated hardware is installed or various programs. It is installed from a network or a recording medium into a general-purpose personal computer or the like capable of performing the above.

As shown in FIG. 16, this recording medium is distributed in order to provide the program to the user separately from the apparatus main body. The program is recorded on the magnetic disk 131, the optical disk 132, and the magneto-optical disk. Thirteen
3 or a package medium including a semiconductor memory 134 and the like, and a ROM 112 in which a program is recorded and a hard disk included in the storage unit 118, which is provided to the user in a state of being pre-installed in the apparatus main body. Etc.

In the present specification, the steps for describing the program recorded on the recording medium are not limited to the processing performed in time series in the order described, but are not necessarily performed in time series. It also includes processing executed in parallel or individually.

In this specification, the system means
A plurality of devices are logically aggregated, and it does not matter whether or not the devices of the respective configurations are in the same housing.

[0098]

As described above, according to the present invention, words can be registered by voice. Further, the registration can be executed without making the user aware of the registration mode. Further, it becomes possible to easily register the unknown word from the continuous input speech including the known word and the unknown word. Furthermore, it becomes possible to reflect the registered word in subsequent dialogues.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of an embodiment of a dialogue system to which the present invention has been applied.

FIG. 2 is a diagram showing an example of a state of a cluster.

FIG. 3 is a diagram showing word registration.

FIG. 4 is a block diagram showing a configuration example of a voice recognition unit in FIG.

5 is a flowchart for explaining the operation of the dialogue system of FIG.

6 is a flowchart for explaining details of the voice recognition process in step S22 of FIG.

7 is a flowchart for explaining details of an operation of generating a word string in step S44 of FIG.

FIG. 8 is a diagram showing an example of a grammar used in a language model database.

FIG. 9 is a diagram showing an example of a language model based on a finite state automaton.

FIG. 10 is a diagram illustrating an example of calculation of a language score using a tri-gram.

FIG. 11 is a diagram showing an example of a tri-gram database.

12 is a flowchart for explaining details of the word acquisition process of step S24 of FIG.

FIG. 13 is a diagram showing an example of a template.

FIG. 14 is a diagram showing an example of a grammar incorporating a phoneme typewriter.

FIG. 15 is a block diagram showing a configuration example of an embodiment of a computer to which the present invention has been applied.

[Explanation of symbols]

1 voice recognition unit, 2 associative storage unit, 3 dialogue control unit, 4 word acquisition unit, 41 microphone, 42
AD conversion unit, 43 feature amount extraction unit, 44 matching unit, 45 phoneme typewriter unit, 46 control unit,
51 acoustic model database, 52 dictionary database, 53 language model database

Claims (11)

[Claims] 1. A voice processing device for processing an input voice and registering a word included in the input voice based on the processing result, the recognition means recognizing continuous input voice, and the recognition means. In the recognition result recognized by the unknown word determination means for determining whether or not an unknown word is included, by the unknown word determination means, if it is determined that the unknown word is included, the unknown word A speech processing apparatus, comprising: an acquisition unit that acquires a word corresponding to the above item; and a registration unit that registers the word acquired by the acquisition unit in association with other information. 2. The pattern determination means for determining whether or not the recognition result matches a specific pattern, wherein the registration means causes the pattern determination means to match the recognition result with a specific pattern. The voice processing device according to claim 1, wherein the word is registered when it is determined that the word is present. 3. The unknown word determination means determines that the unknown word is not included, or the pattern determination means determines that the recognition result does not match a specific pattern. In this case, the voice processing device according to claim 2, further comprising a response generation unit that generates a response corresponding to the input voice. 4. The registration means, as the other information,
The voice processing device according to claim 2, wherein the category of the word is registered. 5. The registration means registers the other information based on the pattern determined to match by the pattern determination means.
The voice processing device according to. 6. The voice processing apparatus according to claim 1, wherein the acquisition unit acquires the word by clustering the unknown word. 7. A comparison unit is further provided which compares acoustic scores of a predetermined section of the input speech matched with a known word and recognized with a phonological typewriter, wherein the comparison unit includes the phonological type. The speech processing apparatus according to claim 1, wherein when the acoustic score recognized by the writer is superior, the section is estimated as an unknown word. 8. The comparing means compares the acoustic score when the known word is matched with the acoustic score when the phonological typewriter is recognized, and then compares the acoustic score. The audio processing device according to claim 7. 9. A voice processing method of a voice processing device, which processes an input voice and registers a word included in the input voice based on a processing result, a recognition step of recognizing continuous input voices, In the recognition result recognized by the processing of the recognition step,
A determination step of determining whether or not an unknown word is included, and, by the processing of the determination step, if it is determined that the unknown word is included, an acquisition step of acquiring a word corresponding to the unknown word; And a registration step of registering the word acquired by the processing of the acquisition step in association with other information. 10. A program of a voice processing device for processing an input voice and registering a word included in the input voice based on a processing result thereof, the recognition step of recognizing continuous input voices, In the recognition result recognized by the processing of the recognition step,
A determination step of determining whether or not an unknown word is included, and, by the processing of the determination step, if it is determined that the unknown word is included, an acquisition step of acquiring a word corresponding to the unknown word; And a registration step of registering the word acquired by the processing of the acquisition step in association with other information, the recording medium having a computer-readable program recorded thereon. 11. A recognition step of recognizing a continuous input voice to a computer for controlling a voice processing device which processes an input voice and registers a word included in the input voice based on the processing result. In the recognition result recognized by the processing of the recognition step,
A determination step of determining whether or not an unknown word is included, and, by the processing of the determination step, if it is determined that the unknown word is included, an acquisition step of acquiring a word corresponding to the unknown word; And a registration step of registering the word acquired by the processing of the acquisition step in association with other information.