JPH11282486A - Sub word type unspecified speaker voice recognition device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform word registration usable by an unspecified speaker with the easiness of the same degree as a specified speaker voice recognition system. SOLUTION: At the time of a word registration mode, an input voice is inputted to a partial word sequence generation part 12 by a mode switching part 11. The partial word sequence generation part 12 converts the inputted voice to the sequence of at least one partial word and registers information corresponding to the partial word sequence to a user registration word dictionary 13. At the time of a voice recognition mode, the input voice is inputted to a main voice recognition part 14 by the mode switching part 11. The main voice recognition part 14 obtains a word voice model in which partial word voice models are connected together from the information corresponding to the respective partial word sequences registered in the user registration word dictionary 13 and recognizes the input voice by using the obtained word voice model.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a subword-type speaker-independent speech recognition apparatus and method that allows a user to easily and appropriately register additional words.

[0002]

2. Description of the Related Art Speech recognition technology plays an important role in realizing an excellent man-machine interface.

[0003] Initially, a specific speaker voice recognition device was used. The specific-speaker voice recognition device utters a word that the user wants to input voice several times before use, and registers a word-speech model for verification in the voice device based on the word voice of the user. When actually inputting voice, the input voice uttered by the user is checked against a word-speech model registered in the device, and the word having the highest matching level is determined as a recognition result (Literature: Masai, Shin Ta, Uehara, "Development of Specific Speaker Word Speech Recognition System Using Differential-Orthogonalization Filter Method", Proc. Of the Autumn Meeting of the Acoustical Society of Japan, 1988, pp.65-66,
October 1988).

[0004] In such a device, since the word voice model inside the device is specialized for the voice of the user at the time of registration, voices other than the registered user cannot be recognized or the recognition performance is significantly reduced. . In order for a different user to use the device, it was necessary to perform a complicated voice registration operation again.

[0005] The specific-speaker speech recognition device is very inconvenient when a plurality of users alternately use the same. In addition, a device such as a vending machine installed on a street cannot register a user's voice, and thus cannot be handled by such a voice recognition device.

For this reason, an unspecified speaker recognition device has been used. In the early unspecified speaker speech recognition device,
First, word voices spoken by a large number of speakers (typically 100 or more) for words to be input to the device are collected. A word voice model is generated from these word voices and registered in the device. When performing voice input,
The input speech emitted by the user is collated with the word speech model registered in the device, and the word having the highest degree of collation is regarded as the recognition result (Reference: Matsuura, Nitta, “SMQ / HMM”
Speaker-Based Vocabulary Word Recognition: Matsuura, Nitta,
IEICE Transactions D-II vol.J76-D-II No.12, p
p.2486-2494, December 1993).

[0007] The word-speech model included in such a device has characteristics that many speakers have in common, and does not depend on the speech of a specific speaker. Therefore, a voice uttered by an unspecified speaker can be recognized.

However, in the initial speaker-independent speech recognition device, it is necessary to collect speech data uttered by a large number of speakers for each word.
There was a problem that the required labor was extremely large.

When the number of speakers for learning the word speech model is relatively small, the word speech model also includes the unique characteristics of the small group (such as intonation and sound deformation that are valid only in a certain specific region / generation). May be generated. In the speech model trained in this way, the recognition performance for an unspecified speaker deteriorates.

In the initial speaker-independent speech recognition device, speech must be collected and processed for each word, and the number of learning speakers per word cannot be so large. There is also a problem that a word-speech model that can be sufficiently matched with the generated speech may not be generated.

Therefore, in recent years, a word-speech model is held in a recognition device in units of a partial-word speech model having acoustically meaning (phonemes and syllables are mainly used as partial words), and a word to be recognized is recognized. Has been developed by connecting partial word speech models to each other, generating a collated partial speech model, and comparing the generated speech model with an input speech (hereinafter, referred to as a subword type unspecified speaker speech recognition method) (Reference:
Mark Punsak, Nitta, “Comparison of Context
Dependent Sub-word HMMs for Japanese ”, IEICE Technical Report vol.93 No.364, pp.63-70, 1993
December).

The configuration of the sub-word type speaker-independent speech recognition apparatus will be described below with reference to FIG.

First, the main speech recognition section 440 comprises an acoustic analysis section 441, a quantization section 442, and an HMM recognition section 443.

The acoustic analyzer 441 analyzes the input audio signal, for example, by LPC (Linear Predictive Coding), and obtains the characteristic parameters of the input audio signal.

The quantization unit 442 converts the characteristic parameters obtained by acoustically analyzing the input speech by the acoustic analysis unit 441 into a label sequence representing a speech segment by statistical quantization.

The HMM recognition unit (HMM collation unit) 443 is
The probability that each of the word HMMs stored in the word HMM dictionary 450 generates a label sequence corresponding to the input speech is calculated, and the word that outputs the label sequence with the maximum probability is output as a recognition result.

Here, the discrete HMM (Hi
dden Markov Model (Hidden Markov Model) will be described.

The HMM is composed of states and transitions, and one label is output when transitioning from a certain state to a certain state. The probability of transition from one state to another state is defined for each state, and the probability that each label is output for each transition is defined.

In practice, the HMM is defined by the following six parameters.

N ^x : Number of states of HMM representing partial word x (states S (1), S (2),..., S (N)) ：: Number of labels (label R = 1, 2,..., Κ) p ^x (i, j): transition probability of HMM representing partial word x (probability of transition from S (i) to S (j)) q ^x (i, j, k): SMM of HMM representing partial word x Probability of outputting label k at the time of transition from (i) to S (j) ^mx (i): Probability of initial state of HMM representing partial word x (probability that S (i) becomes initial state) F ^x : a set of states that can be the final state of the HMM representing the partial word x The HMM has a transition restriction reflecting the characteristics of speech. In speech, in general, a loop transition that returns from the state S (i) to the states S (i-1) and S (i-2) previously passed is not allowed because the temporal context is disturbed. FIG. 45 shows three states 2
3 shows a discrete HMM of a loop. Here, the final state S (N),
That is, S (3) does not contribute to collation.

As described above, the collation between the HMM and the label sequence is performed by calculating the probability or the logarithm of the probability that the HMM outputs the label sequence. However, in an actual apparatus, Viterbi ( viterbi) A value calculated by an algorithm (Viterbi score) is often substituted.

The Viterbi score is a logarithmic value of a probability that a label sequence is output when a state transition occurs so as to output an input label sequence with the highest probability.

The input label sequence is represented by Y = y (1), y (2),
.., Y (L), the Viterbi score can be calculated as follows.

To calculate the Viterbi score, the array D
Use (T, M).

(1) D (0,1 to N), that is, D (0,1) to D
Initialize (0, N) with the logarithmic value of the initial state probability. That is, D
Let (0,1) = ln m _{1 to} D (0, N) = ln m _N. In the case of the HMM having the configuration shown in the figure, D (0,1) = 0, D (0,2 to N) =
−∞ Further, it is initialized to D (0 to T, 0) = − ∞.

(2) Steps (3) and (4) are repeated while increasing t by 1 from 1 to T.

(3) Repeat (4) while increasing n by 1 from 1 to N.

(4) d ₁ = D (t-1, n-1) + ln p (n-
1, n) + ln q and (n-1, n, y (t)), d 2 = D (t-1, n) + ln
Calculate p (n, n) + lnq (n, n, y (t)) and substitute the larger value for D (t, n).

(5) The desired Viterbi score can be obtained from D (T, N).

The partial word HMM is a discrete HMM created for each partial word. Here, it is assumed that a phoneme is used as a unit of a partial word, and that the model is modeled by a discrete HMM having two loops and three states.

The partial word HMM is a partial word HMM dictionary 46
0 is registered. An example of the storage format (registration format) of the partial word HMM registered in the partial word HMM dictionary 460 is shown in FIG. In this example, (the parameter of) the partial word HMM is registered as a pair with the partial word model name.

The word HMM is a partial word HMM dictionary 460
Can be created by linking the partial words HMM registered in the. At this time, the final state S (N) of each partial word HMM is superimposed on S (1) of the succeeding partial word HMM connected immediately thereafter. For example, when the word "adult" is represented by a partial word, "o, t, o, n, a"
Therefore, the corresponding word HMM is as shown in FIG.

The word HMM is registered in the word HMM dictionary 450. Word H registered in this word HMM dictionary 450
FIG. 48 shows an example of the storage format of the MM. In this example, the parameters of the word HMM configured as described above are
It is stored in pairs with word names.

It should be noted that the word HMM is not composed for each word and its parameters are stored, but the partial word HMM names constituting the word are stored, and the partial word HMM
There is also a device configuration in which a word HMM is constructed with reference to the dictionary 460 and then collated. Furthermore, readings written in hiragana or the like are stored, and are converted to partial word names during collation,
There is also a device configuration in which a word HMM is constructed with reference to the partial word HMM dictionary 460 and then collation is performed.

The HMM recognizing unit 443 stores the word HMM dictionary 45
The Viterbi score for the input label sequence is calculated for each word using the respective parameters of the word HMM registered in 0. Then, a word having the largest Viterbi score is output as a recognition result.

In the present system, by inputting the reading of a word, words that can be input by voice can be added or changed. Therefore, compared with the initial unspecified speaker voice recognition method,
The trouble is greatly reduced.

Further, since the necessary partial word speech model can be used in common between devices having different recognition vocabularies, it can be generated from speech data uttered by a large number of learning speakers. Therefore, a word voice model more suitable for recognizing a voice uttered by an unspecified speaker can be generated.

[0038]

In the above-mentioned sub-word type speaker-independent speech recognition system, registration and recognition are performed using recognition words.
When changing, the user has to input a phoneme sequence or a hiragana string representing the reading of a word, but this is difficult if there is no specialized knowledge about character input means and voice.

That is, the registration operator needs to be familiar with the correspondence between the reading method (speech method) desired to be registered in the speech recognition device and the symbol expressing the reading method. When words are read in hiragana, the correspondence is relatively easy to understand, but even in this case, the presence or absence of prolonged sound (the hiragana string "tokei" is / to
Does it correspond to the utterance of "kei /" or "/ tok"
Knowledge of how the device interprets e./e./ etc.) is still needed.

In the case of a portable device or a vehicle-mounted device,
It may be difficult to attach a character input device. Even in such a case, it is possible to input the reading method by using a method of displaying the whole phoneme (or hiragana) on the screen and selecting one character at a time, a method using character recognition technology, or the like. Requires complicated operations and is not always convenient.

On the other hand, in the specific speaker recognition system, it is possible to register a user-specific word only by saying a word to be registered several times. In other words, there is no need to be familiar with the correspondence between phonological symbols, hiragana and pronunciation, and no character input means is required. Therefore, addition and change of the recognition word could be easily performed. However, there is a drawback in that the user must first register all words to be recognized after obtaining the device. On the other hand, in a device using the unspecified speaker recognition method, it is possible to incorporate words considered to be frequently used for input into the device in advance, so that the user wants to make a special name by himself / herself. The device can be started to be used only by registering words only for words.

Therefore, in order to eliminate the drawbacks of the above two methods, collation is performed by using unspecified speaker recognition and specific speaker recognition at the same time on the input speech, and the likelihood ( There is a method in which the one with the higher degree of certainty of recognition) is used as the final recognition result. In this method, words that are considered to be frequently used for input are registered in advance in the unspecified speaker recognition unit at the stage of designing and manufacturing the device, and words added or changed by the user are specified speakers. Register in the recognition unit. With such a configuration, the user only needs to register words that he or she wants to add or change, and does not need to input symbols at the time of registration.

However, since both the unspecified speaker recognition method and the specific speaker recognition method must be incorporated in the apparatus,
The device becomes complicated. In addition, since different methods are used in combination, the respective likelihood scales are different, and a correction is required to compare the two. However, it is difficult to determine a correction method that can always be used. Furthermore, the words registered for specific speaker recognition strongly depend on the voice of the registrant, and the disadvantage of the specific speaker recognition method in which a person other than the registrant cannot correctly verify when the same word is uttered is still overcome. Can not.

The dictionary used in the sub-word type speaker-independent speech recognition apparatus generates words by connecting general partial word models in accordance with reading information represented by phonemes, phonemes, and the like. When pronunciation of some partial words is unusual due to dullness or the like, the accuracy of speech recognition is reduced. Therefore, there is a problem that a user having a dullness or the like is forced to always use a speech recognition device having low recognition accuracy (for the speaker).

The present invention has been made in consideration of the above-described circumstances, and has as its object to provide a subword-type unspecified speech that can perform word registration as easily as a specific speaker speech recognition method. It is an object of the present invention to provide a person voice recognition apparatus and method.

Another object of the present invention is to provide a subword-type speaker-independent speech recognition apparatus and method capable of updating a word dictionary according to pronunciation of a user.

Still another object of the present invention is to provide an apparatus and method for recognizing a subword-type unspecified speaker that can prevent erroneous registration in registration of a word dictionary according to pronunciation of a user. is there.

Still another object of the present invention is to provide an apparatus and method for recognizing a subword-type unspecified speaker whose structure can be simplified.

It is still another object of the present invention to provide a subword-type speaker-independent speech recognition apparatus and method capable of sequentially adapting to a user's speech to improve recognition accuracy.

[0050]

According to a first aspect of the present invention, there is provided a partial word sequence generating means for converting an input speech into a sequence of at least one partial word, and a partial word sequence generating means. The user-registered word dictionary in which information corresponding to the partial word sequence converted by the above is registered, and the partial-word speech models are connected from the information corresponding to each partial word sequence registered in the user-registered word dictionary. A word voice model obtaining means for obtaining a word voice model; and a main voice recognition means for recognizing a voice uttered by the user using the word voice model obtained from the user registered word dictionary. And

In such a configuration, the input speech is not directly converted into a word-speech model and registered while using the input by voice without using the input by character information at the time of word registration. Recognizes the voice uttered by the user (during speech recognition mode) by converting it to a partial word sequence such as a phoneme and registering the information corresponding to the partial word sequence in a dictionary (registering it in the user registration word dictionary) A word speech model in which partial word speech models for unspecified speaker recognition are joined from information corresponding to a partial word sequence in the dictionary, and perform speech recognition using the word speech model. Therefore, the registered word can be used by an unspecified speaker despite the fact that the registered word has been registered by voice. On the other hand, in the prior art, when word registration by voice is applied, the registered words are exclusively used for the registered speaker, and the recognition performance when another speaker uses the word is extremely high. bad. In addition, when word registration using character information is applied, an input means for character information is required, and the operation is complicated.

Here, as the information corresponding to the partial word sequence registered in the user registration word dictionary, the partial word sequence itself or the partial word speech model corresponding to each partial word constituting the partial word sequence is stored in advance. Any of the word voice models created by joining together may be used.

In the former case, the means for acquiring the word speech model (word speech model acquisition means) connects the corresponding partial word speech model from the information (partial word series itself) corresponding to the partial word series. It is necessary to have a function of acquiring the word voice model by creating the word voice model. On the other hand, in the latter case, the word-speech model acquisition means only has a function of extracting information corresponding to the partial word series from the user registered word dictionary,
It is possible to acquire a word voice model. However, when generating information corresponding to the partial word sequence from the partial word sequence converted from the input voice, it is necessary to create a word voice model by connecting the corresponding partial word voice models from the partial word sequence. .

The configuration according to the second aspect of the present invention is the same as the configuration according to the first aspect, except that the partial word sequence converted by the partial word sequence generating means satisfies a predetermined registration condition. Is added, and only the partial word sequence determined to satisfy the registration condition is registered in the user registered word dictionary.

In such a configuration, by appropriately setting the registration condition, the partial word sequence converted and output as a result of the recognition error in the partial word sequence generation means, that is, the partial word sequence which does not clearly correspond to the input speech It is possible to prevent the sequence from being registered in the user registration word dictionary and causing the recognition performance of the main voice recognition unit to be reduced.

Here, as the registration condition, for example, a condition for determining whether or not the condition is satisfied by comparison with the likelihood (recognition probability) of the partial word sequence converted (generated) by the partial word sequence generating means is applied. It is possible. in this way,
By examining the likelihood of the partial word sequence and not registering the partial word sequence if it is equal to or less than the reference value, the partial word sequence generation means is generated when the plausible partial word sequence cannot be output (misrecognized). The problem that a sequence is registered can be reduced.

In addition, an upper limit N is set for the number of partial word sequences that can be registered per word, and the upper limit N is used as one of the registration conditions. Are registered in the user registration word dictionary irrespective of the likelihood. If the number exceeds N, the top N
Individuals can be registered in the user registration word dictionary.
In this way, in an apparatus having a large memory (storage area) restriction, the number of registered partial word sequences can be limited.

The structure according to the third aspect of the present invention uses the information according to the structure according to the first aspect, for all the partial word sequences converted by the partial word sequence generating means, using information representing the partial word sequences. To the user, accepts the designation from the user as to whether or not the registration is possible, and adds a registration confirmation means to confirm whether or not the corresponding partial word sequence can be registered according to the accepted specification, and registers from the user. Only information corresponding to the instructed partial word sequence is registered in the user registration word dictionary.

In such a configuration, even if a recognition error occurs in the partial word sequence generation means due to the influence of noise or the like, even if a large value likelihood is obtained by chance depending on the type of noise, the value can be obtained. By presenting the information indicating the partial word sequence to the user, the user can confirm the recognition error in the partial word sequence generation means and can instruct whether or not registration is to be performed. It can be prevented from being registered in the user registration word dictionary.

The configuration according to the fourth aspect of the present invention uses the information according to the configuration according to the first aspect, for all the partial word sequences converted by the partial word sequence generating means, using information representing the partial word sequences. To the user, accepts the user's editing operation for the information, performs the editing process on the information, reflects the result of the editing process on the corresponding partial word sequence, and determines whether the information can be registered. A registration / editing unit that accepts a specification from the user and checks whether or not a corresponding partial word sequence can be registered in accordance with the received specification, and a part generated from the input speech by the partial word sequence generation unit. In addition to enabling word series correction, only information corresponding to the partial word series for which registration has been instructed by the user has been registered in the user registration word dictionary. Characterized in that was.

In such a configuration, if there is a recognition error in the partial word sequence generation means, it can be registered after confirming and correcting it, so that registration can be made without re-uttering.

According to a fifth aspect of the present invention, there is provided the configuration according to any of the first to fourth aspects, wherein information corresponding to the partial word sequence registered in the user registration word dictionary is written in characters. User word registration dictionary display means for converting the information into information and presenting it to the user is added.

In such a configuration, the user can check the information registered in the user registered word dictionary at a later date.

According to a sixth aspect of the present invention, there is provided the configuration according to any of the first to fourth aspects, wherein information corresponding to the partial word sequence registered in the user registration word dictionary is written in characters. A user who converts the information into information and presents it to a user, accepts a user's editing operation on the information, performs an editing process on the information, and reflects a result of the editing process on a user registration word dictionary. A feature is that a word registration dictionary editing means is added.

In such a configuration, the user can check the information registered in the user registration word dictionary, and can correct any troubles.

According to a seventh aspect of the present invention, there is provided the configuration according to any one of the first to fourth aspects, wherein the information corresponding to the partial word series generated from the character string information representing the reading of the word is provided. Adds a character registered word dictionary registered in the same expression format as the user registered word dictionary, and corresponds to each partial word sequence registered in the user registered word dictionary during recognition processing by the main voice recognition means. The word speech model in which the partial word speech models are joined from the information to be obtained is obtained, and the word speech model in which the partial word speech models are joined is also obtained from the information corresponding to each partial word sequence registered in the character registration word dictionary. It is characterized in that a voice uttered by the user is recognized using the acquired word voice models.

In such a configuration, although the user registration word dictionary and the character registration word dictionary are registered in different ways, the expression format (registration format) of the information corresponding to the partial word sequence By matching
It is possible to perform voice recognition by using both of them at the same time with only one recognition method, and the configuration of the device can be simplified. Here, the character registration word dictionary may be one that is pre-installed in the device in a state where recognition words are registered in advance, or one that is provided in a state stored in a removable recording medium. Absent. Also, the information may be registered in the same device based on word reading information input from character input means such as a keyboard.

The configuration according to the eighth aspect of the present invention is characterized in that the partial word sequence generating means in any one of the configurations according to the first to fourth aspects can be used not only in the word registration mode but also in the speech recognition mode. In addition to the configuration in which a voice uttered by the user is recognized to generate at least one partial word sequence, the following in-use word registration determining means, that is, in the voice recognition mode, the partial word sequence generating means Whether the partial word sequence generated by the partial word sequence generation unit can be registered based on at least one of the likelihood of the generated partial word sequence, the recognition result of the main speech recognition unit, and the likelihood of the recognition result And a new word-in-use determining means for additionally registering the information of the partial word series in the user-registered word dictionary according to the result of the determination is provided.

In such a configuration, when the user makes a special utterance (for example, strong rounding), the recognition accuracy of the main voice recognition means tends to decrease, and the likelihood of the recognition result tends to decrease. However, the information of the partial word sequence generated by the recognition processing by the partial word sequence generation means performed in parallel with the recognition processing by the main voice recognition means is stored in the user registration word dictionary by the use word registration determination means. By registering and using it again from the next time, recognition accuracy of the main voice recognition means can be improved.

Here, in order to determine whether a partial word sequence can be registered or not, a first method in which only a partial word sequence having a likelihood larger than a reference value can be registered, the likelihood of a partial word sequence is recognized A second method in which only a partial word sequence having a likelihood greater than a reference value by a reference value or more than the likelihood of a recognition result can be registered in comparison with the likelihood of the result. A third method in which a partial word sequence or a partial word sequence having a certain number of higher likelihoods as an upper limit can be registered. A word uttered by a user is known, and the word does not match the recognition result. In the fourth method, a partial word sequence generated by the recognition processing in the partial word sequence generating means can be registered, and the likelihood of the partial word sequence to be added to the fourth method is added to the determination condition. If the word uttered by the user does not match the recognition result, Including a fifth scheme likelihood of word sequence is to be registered is greater than the reference value can be applied.
In order to make the above fourth or fifth method applicable, a special mode (adaptive mode) is prepared and
In this mode, an interface function may be provided in which a word is presented from the device to the user (the presented word may be specified by the user) and the word is uttered by the user.

As a result, it is possible to realize a speech recognition device that sequentially adapts to the user's speech.

The configuration according to the ninth aspect of the present invention is characterized in that the partial word sequence generation means in any of the configurations according to the first to fourth aspects does not only operate in the word registration mode but also in the speech recognition mode. In addition to the configuration of generating a sequence of at least one partial word by recognizing a voice uttered by a user, the user uses the following word registration confirmation means at the time of use, that is, when the recognition result is output by the main voice recognition means. , And when the registration instruction is received, information generated by the partial word sequence generation unit and corresponding to the registered partial word sequence is added to the user registration word dictionary. It is characterized in that a use-time word registration confirmation means for registration is newly provided.

In such a configuration, since the user makes a special utterance (eg, strong rounding), the recognition accuracy of the main voice recognition means is reduced, the likelihood of the recognition result is reduced, and the recognition result is reduced. Even in the case of an error, the information of the partial word sequence generated by the recognition processing by the partial word sequence generation means performed in parallel with the recognition processing by the main voice recognition means can be used as a result of the recognition of the main speech recognition means by the user. Is registered in the user registration word dictionary in accordance with the registration instruction in consideration of the above, and the recognition can be performed using the same from the next time, whereby the recognition accuracy of the main voice recognition means can be improved. As a result, it is possible to realize a speech recognition device that sequentially adapts to the user's speech.

It should be noted that, instead of the partial word sequence generation means in the configuration according to the eighth or ninth aspect, in the voice recognition mode, only when a predetermined condition is satisfied,
In addition to using a partial word sequence for performing a recognition process on an input voice to generate a partial word sequence, input voice storage means for temporarily storing a voice uttered by a user in a voice recognition mode is newly provided. Instead of the word registration determination means, in the voice recognition mode, the presence or absence of the above condition is determined based on the likelihood of the recognition result of the main voice recognition means. The stored speech is input to the partial word sequence generation unit to operate the partial word sequence generation unit, the likelihood of the partial word sequence generated by the partial word sequence generation unit, the recognition result of the main speech recognition unit, And determining whether or not to register the partial word sequence generated by the partial word sequence generation means based on at least one of the likelihoods of the recognition result and the partial unit according to the determination result. It is also possible to adopt a configuration in which the information of the sequence provided when using the word registration decision means for additionally registered in the user registration word dictionary.

In such a configuration, when the likelihood of the recognition result of the main speech recognition means does not satisfy the above condition, for example, when the likelihood of the recognition result is large, the recognition processing by the partial word sequence generation means is not executed. Since the partial word sequence generation unit is not activated for an average utterance by the user, it is suitable for realizing the function of the partial word sequence generation unit on a computer that is not fast.

[0076]

Embodiments of the present invention will be described below with reference to the drawings.

[First Embodiment] FIG. 1 is a block diagram showing a sub-word type speaker-independent speech recognition apparatus according to a first embodiment of the present invention.

In the apparatus shown in FIG. 1, two modes of word registration and voice recognition (recognition processing) can be selected and designated by the user. At the time of word registration, an input voice is input to the partial word sequence generation unit 12 by switching of the mode switching unit 11, and the input voice is converted into a sequence of at least one partial word corresponding to a phoneme, a phoneme, or the like. Information corresponding to the partial word sequence is registered in the user registration word dictionary 13. On the other hand, at the time of recognition processing, the input voice is input to the main voice recognition unit 14 by the mode switching unit 11 and acquired from the registered contents of the user registration word dictionary 13, and a word voice model in which partial word voice models are joined, for example, The input voice is recognized using the parameters of the word voice HMM in which the parameters of the partial word HMM are connected.

As described above, at the time of word registration, the input speech is first input to the partial word sequence generation unit 12. The partial word sequence generation unit 12 converts the input voice into a partial word sequence.

FIG. 2 shows an example of the internal configuration of the partial word sequence generation unit 12.

Here, the partial word sequence generation unit 12 includes an acoustic analysis unit 121, a quantization unit 122, a partial word connection table 12
3. It is composed of a partial word HMM recognition unit 124, a partial word HMM dictionary 125, and a word HMM generation unit 126.
The sound analysis unit 121 and the quantization unit 122 are the same as the sound analysis unit 441 and the quantization unit 442 used in the main speech recognition unit 440 of the conventional subword type unspecified speaker speech recognition device shown in FIG. Also, the partial word HMM dictionary 125
Corresponds to the partial word HMM dictionary 460 in FIG.

In the partial word connection table 123, combinations of phonemes that can be directly connected are registered. This table (table) 123 includes “consonants and consonants do not connect”, “consonants,
It is used for recognition with higher accuracy by using a restriction of Japanese speech such as "the repellent does not exist at the beginning of the word." FIG. 3 shows an example of the partial word connection table 123. In this example, a phoneme that can follow for each preceding partial word is registered. In FIG. 3, a partial word “#” is a symbol for a virtual phoneme indicating a head, and “&” is a symbol for a virtual phoneme indicating an end.

The partial word HMM recognizing section 124 outputs the partial word H
Using the MM dictionary 125, the input speech is recognized in partial word units (here, phoneme units), and at least one partial word sequence in which partial words are connected according to the partial word connection table 123 is output as the recognition result. . This partial word H
Details of the MM recognizing unit 124 will be described below with reference to flowcharts of FIGS.

There are countless series formed by connecting partial words according to the partial word connection table 123. The partial word HMM recognizing unit 124 searches for a partial word model sequence that outputs a label sequence corresponding to the input speech and its Viterbi score while dynamically generating the sequence.

This is calculated as follows.

The sequence of partial words is represented by Χ = [x (1), x (2),
, X (J (X))]. J (X) is the length of the partial word sequence. It holds an array D used in Viterbi algorithm for each sequence of partial word used in matching them to be represented as D ^X. Further, to hold the number of states the direction of the size of the array D ^X to Eta ^X. Η ^X is a partial word x (J
It is equal to the number of states Nx ^{(J (X))} of the partial word HMM corresponding to ^(X)) . However, the number of states for the virtual phoneme “#” is one. That is, Η ^[#] = 1. Sequence Z holds D ^X.

First, an array D ^[#] is generated, and D ^[#] (1,
1) Initialize to = 0 and add D ^[#] to Z (steps S1 to S3).

Next, while increasing t by 1 from 1 to T (steps S4, S5, S21), steps S6 to
S20 is repeated.

[0089] At step S6～S20, while removed D ^X from Z in ascending order of one by one sequence length J (X) (step S6, S7), step S8~S20 is repeated.

In step S8, it is checked whether or not X is [#]. When X = [#], if t = 1, 0 is substituted into D ^X (t, 1), If t ≠ 1, D ^X (t, 1)
Is substituted into (step S9), and the process proceeds from step S18.

On the other hand, when X ≠ [#], processing of the first state of the partial word described below (step S10,
S11, S12 or S10, S11, S13) are performed.

[0092] First of all, the history of D ^X of interest current X =
The history obtained by removing the last partial word of [x (1), x (2), ..., x (J (X))] is X1 = [x (1), x (2), ..., x (J
(X) -1)] (step S10).

Next, it is checked whether or not D ^X1 and H ^X1 exist in Z (step S11), and if they exist, they are taken out and step S12 is performed. If not, step S13 is performed.

In step S12, d ₁ = D ^X1 (t,
And _{^{Η X1), d 2 = D}} X (t-1,1) + ln p (1,1) + ln q
The larger value of (1,1, y (t)) is substituted for D ^X (t, 1).

On the other hand, in step S13, d = D
^{X (t-1,1) + ln} p (1,1) + ln q to calculate the (1,1, y (t)) , is substituted for D ^X (t, 1).

When step S12 or S13 is executed, the second and subsequent states of the partial words described below (second states to
The processing of the N ^{J (X)} state) (steps S14 to S17) is performed.

Here, while increasing n by one from 2 to N ^{J (X)} (steps S15 and S17), step n
Repeat step 16. In this step S16, d ₁ = D ^X
(t-1, n-1) + lnp (n-1, n) + lnq (n-1, n, y (t)),
d ₂ = D ^X (t-1, n) + ln p (n, n) + ln q (n, n, y (t))
Is calculated, and the larger value of d ₁ and d ₂ is substituted for D ^X (t, n).

[0098] ^{Then, D X (t, Η X} ) whether is examined is ^{-∞, D X (t, Η} X) returns to step S6 without nothing if the -∞. On the other hand, if D ^X (t, Η ^X ) is not −∞, a new D generation process (steps S19 and S20) described below is executed.

Here, all the partial words x ′ ₁ , x ′ ₂ ,... That can follow the partial word x (J (X)) at the end of X are searched from the partial word connection table 123, connect a new sequence X _'1, X' _2, making ... (step S1
9).

That is, X ′ ₁ = [x (1), x (2),.
x (J (X)), x ′ ₁ ], X ′ ₂ = [x (1), x (2),.
x (J (X)), x ′ ₂ ],.

Next, the sequence X 'generated in step S19
_1, X _'2, ... for each, D the corresponding
It is checked whether or not already exists in Z. If not, D and Ｄ are newly generated and added to Z (step S20). Here, all newly created Ds are D (0 to T,
1 to N) = -∞.

[0102] By executing the step S20, the processing of D ^X that is currently attention will end, and the flow returns to step S6.

The above operation was repeated while increasing t by 1 from 1 to T, and by taking out D ^X from Z for each t one by one in order of decreasing sequence length J (X). When the value exceeds T, the D included in Z
^X (T, ^{Η X)} so that the Viterbi score is been determined for each series Χ to. Therefore, by referring to the partial word connection table 123, a partial word sequence in which the partial word at the end of the sequence X can be connected to "&" is selected and sorted in descending order, so that the sequence ( A set of (partial word series) X and its Viterbi score can be obtained.

Actually, when T increases, the number of Ds held in Z increases explosively. Therefore, the generation and calculation of D are limited to be performed only under certain conditions, and the speed is increased. Often attempts.

In brief, instead of comparing with -∞ at the step S18, f (t) = αt (where α is a constant)
To limit the creation of new D,
Higher speed is achieved by suppressing the number of D in Z. Further, when returning to step S6, the noted D ^X (t, 1 to
N) is, when g (t) = βt (where β is a constant) All than smaller, by removing the D ^X from Z, is also performed to reduce the D contained in Z.

If the speed-up method is not performed at all, the processing in step S13 is not necessary. However, when speeding up using, for example, g (t), D ^X1 referred to in step S11 may be deleted. Therefore, step S13 is required.

There are various other calculation methods and speeding-up methods for obtaining the partial word sequence, but the present invention does not care about the recognition method for this part.

By the way, the partial word HMM recognizing section 124 determines if there is a partial word sequence whose Viterbi score for the label sequence input as described above is larger than the function g (T) = γT of the input label sequence length T. All of them are selected, and if they do not exist, a partial word sequence having the maximum Viterbi score is output.

As a method of selecting a partial word sequence to be output, only a sequence having the maximum Viterbi score is output, or a predetermined number of higher-order sequences are output.
There are also other methods.

One or more partial word sequences output from partial word HMM recognizing section 124 are passed to word HMM generating section 126. The word HMM generation unit 126 refers to the partial word HMM dictionary 125 based on the partial words constituting the partial word sequence output from the partial word HMM recognition unit 124. The registered contents of the partial word HMM dictionary 125 are as follows:
This is the same as the registration content of the partial word HMM dictionary 125 (see FIG. 46) of the conventional subword type unspecified speaker voice recognition apparatus shown in FIG. 44, and the partial word HMM dictionary 125 includes various partial word speech models. HMM as a partial word
Are registered in pairs with partial word names (partial word model names).

Thus, the word HMM generating unit 126 connects the partial words HMM (parameters) registered in the partial word HMM dictionary 125 according to the partial word sequence output from the partial word HMM recognizing unit 124, Word HM as a word speech model for words uttered by the user
Generate M (parameters of M). Then, the word HMM generation unit 126 registers the pair of the word name of the word uttered by the user and the corresponding word HMM (parameter thereof) as the information of the partial word sequence output from the partial word HMM recognition unit 124. Register in the word dictionary 13.

An example of the user registered word dictionary 13 is shown in FIG. FIG. 7 shows a partial word HMM in the partial word sequence generation unit 12 when the user utters “shain” for the word “employee” during word registration (word registration mode).
The recognition unit 124 outputs one partial word sequence “y, a, i, N”, and when the user utters “Yakuin” for the word “officer”, the partial word sequence generation unit 12
The partial word HMM recognizing unit 124 in “y, a, k, u,
An example of registration of a pair of a word name and a word HMM when two partial word sequences of "i, N" and "y, a, p, u, i, N" are output is shown.

On the other hand, during the recognition process (voice recognition mode), the input voice is input to the main voice recognition unit 1 by the mode switching unit 11.
4 is input.

The main speech recognition section 14 is a user registered word dictionary (corresponding to the word HMM dictionary 450 in FIG. 44) in exactly the same manner as the conventional subword type unspecified speaker speech recognition apparatus shown in FIG. The Viterbi score for the input label sequence is calculated for each word using the respective parameters of the word HMM registered in No. 13. And the main voice recognition unit 14
Outputs the word having the largest Viterbi score as a recognition result.

The configuration of the main speech recognition section 14 is the same as that of the conventional subword type unspecified speaker speech recognition apparatus, and as shown in FIG. 8, (the acoustic analysis section 441 and the quantization section 4 in FIG. 44).
42, and an HMM recognition unit 443). The sound analysis unit 141, the quantization unit 142, and the HMM recognition unit 143 are included. Here, the sound analysis unit 1 in the main voice recognition unit 14
It is not necessary to independently provide the sound analysis unit 121 and the quantization unit 122 in the partial word sequence generation unit 12 and the quantization unit 142 and the partial word sequence generation unit 12. Can be.

In the example of the user-registered word dictionary 13 shown in FIG. 7, when the user inputs "shain" by voice during the recognition processing, the main voice recognition unit 14 generates a label sequence generated from this voice. On the other hand, the Viterbi score of the employee's word HMM and the officer's word HMM (there are two) is calculated.

If the "Employee" has a Viterbi score of -40,
Assuming that the Viterbi score of “executive” is −80 and −100, the result of recognition by the main voice recognition unit 14 is the word “employee”.

When the user inputs "Yakuin" by voice, a Viterbi score is similarly calculated for the label sequence generated from this voice, and the Viterbi score of "employee" is -50, "Officer's Viterbi score is -30 and-
If it is 40, the recognition result is the word "officer".

In the above example, the description has been made assuming that the parameters of the word HMM are directly registered in the user registration word dictionary 13.
The partial word sequence output from the recognition unit 124 may be registered in the user registered word dictionary 13 as shown in FIG.

When a format as shown in FIG. 9 (instead of FIG. 7) is applied as the dictionary registration format (dictionary structure) of the user registration word dictionary 13, the partial word sequence generation unit 12 outputs a partial word sequence. It is not necessary to generate a word HMM because it is only necessary to register in the user registration word dictionary 13. Therefore, unlike FIG.
There is no need to provide the M generator 126. FIG. 10 shows the configuration of the partial word sequence generation unit 12 in this case.

On the other hand, the main voice recognition unit 14
Unlike FIG. 11, the partial word HMM dictionary 145 and the word HMM generation unit 1 (corresponding to the partial word HMM dictionary 125 and the word HMM generation unit 126 in FIG. 2)
46 need to be added. Word HMM generation unit 146
Refers to the user registered word dictionary 13 to obtain a partial word sequence of each word, and refers to the partial word HMM dictionary 145 with the partial words constituting the partial word sequence to obtain a partial word HMM ( Parameter), and
By concatenating them, a word HMM of each word is generated.

The HMM recognizing section 143 calculates the Viterbi score of the word HMM of each word generated by the word HMM generating section 146, and outputs the word having the maximum Viterbi score as a recognition result.

When the registration format of the user registration word dictionary 13 is as shown in FIG. 9 and the configuration of FIGS. 10 and 11 is applied, the partial word sequence generation unit 12 having the configuration of FIG. The HMM dictionary (125) and the partial word HMM dictionary (14
If the contents of 5) are the same, one of them may be shared and the other may not be necessary. In this case, the partial word HMM dictionary to be shared and used may be provided outside the partial word sequence generation unit 12 and the main speech recognition unit 14. Also, the word HMM generation unit (126 or 146) may not be a part of the partial word sequence generation unit 12 or the main speech recognition unit 14, but may be provided outside.

When the registration format of the user registration word dictionary 13 is as shown in FIG. 9, the partial word HMM used in the main speech recognition unit 14 having the configuration shown in FIG. 11 and the partial word sequence generation having the configuration shown in FIG. The partial word HMM used in the unit 12 may be different. For example, the partial word sequence generation unit 12
When a partial word sequence is generated by using a five-state partial word HMM to perform high-precision matching, the main-speech recognition unit 14 uses three-state partial-word HMMs to perform high-speed matching with a large number of words.
It is also possible to apply a configuration that uses the partial word HMM of the state. As described above, when the partial word HMM used in the main speech recognition unit 14 is different from the partial word HMM used in the partial word sequence generation unit 12, FIG.
As in the example of FIG. 11 and the example of FIG. 11, the respective partial word HMM dictionaries (125, 145) may be separately prepared.

The system of the partial words used in the main speech recognition unit 14 and the system of the partial words used in the partial word sequence generation unit 12 may be different. For example, the main speech recognition unit 14 may use a phoneme distinguished for each immediately preceding phoneme as a partial word. That is, “a” after the phoneme “k” is set to “(k) a”, and “a” after the phoneme “s” is set to “(s) a” to distinguish them.

For this purpose, as shown in FIG. 12, it is necessary to provide a partial word HMM dictionary 15 to which the system of partial words used in the main speech recognition unit 14 is applied, and a word HMM generation unit 16.

The registration format of the partial word HMM dictionary 15 itself is the same as that of the partial word HMM described above except that a phoneme distinguished for each immediately preceding phoneme is used as a partial word.
Dictionaries 125 and 144, and a partial word HMM dictionary 460
(See FIG. 44). FIG. 13 shows an example of the partial word HMM dictionary 15. “#” In FIG. 13 is a symbol given to a virtual phoneme representing the beginning of a word.

In the configuration shown in FIG. 12, the word HMM generation unit 16 converts the partial word sequence (phoneme sequence) generated by the partial word sequence generation unit 12 into a system of partial words used by the main speech recognition unit 14. Then, the word HMM generation unit 16 selects a partial word HMM registered in the partial word HMM dictionary 15 in accordance with the converted partial word sequence, connects it, creates a word HMM, and creates a user registered word. Register in the dictionary 13.

For example, assume that when the user utters “employee” for the word “employee” and registers it, the partial word sequence generation unit 12 outputs the sequence “y, a, i, N”. Think.

In this case, the word HMM generation unit 16 converts each phoneme of the partial word sequence “y, a, i, N” into the partial word sequence “(#) y, (y) a , (A)
i, (i) N ". Further, "(#) y, (y)
a, (a) i, (i) N "and the partial words H
The corresponding partial word HMM is extracted from the MM dictionary 15,
A word HMM is generated by concatenating them.

The operations of the partial word sequence generation unit 12 and the main speech recognition unit 14 are the same as those in the configuration of FIG.

[Second Embodiment] Next, a second embodiment of the present invention will be described.

First, consider the case where the user utters “Shain” for the word “employee” and uses it for registration. At the time of registration of the voice, if noise is mixed in a portion immediately before the voice, a partial word sequence including the noise portion is generated. That is, "p, a, h, u, sy, a, i,
A partial word sequence such as "N" is generated. Here, the part “p, a, h, u” is a partial word sequence that is erroneously generated for noise.

The user registration word dictionary (1)
If registered in 3), the word HMM of the word "employee" for the voice "shain" uttered by the user during the recognition process
("P, a, h, u, sy, a, i, N") has a small Viterbi score. Therefore, it is difficult to select “employee” as the recognition result, and the recognition performance is reduced.

The second embodiment realizes a mechanism for automatically preventing the registration of such an erroneous partial word sequence.

FIG. 14 is a block diagram showing a sub-word type speaker-independent speech recognition apparatus according to a second embodiment of the present invention. The same parts as those in FIG. 1 are denoted by the same reference numerals.

In the configuration shown in FIG. 14, partial word sequence generation section 12 converts an input speech into one or a plurality of partial word sequences (at least one partial word sequence) and outputs the partial word sequence. At this time, the partial word sequence generation unit 12
Is different from the configuration of FIG. 10 in the first embodiment, and outputs the Viterbi score of the partial word sequence simultaneously with the partial word sequence.

A feature of the configuration shown in FIG. 14 is that a registration condition determining unit 21 is added to the configuration shown in FIG. The registration condition determination unit 21 receives the partial word sequence output from the partial word sequence generation unit 12 and the Viterbi score of the sequence.

The registration condition judging unit 21 calculates the Viterbi score sent from the partial word sequence generation unit 12 in pair with the partial word sequence by a registration judgment function γ which is a function of the label sequence length T.
(T) = RT (R is a constant) and γ (T) = R
Only when the score is larger than the value of T, the corresponding partial word sequence is registered in the user registration word dictionary 13.

Since the partial word HMM is configured so that the Viterbi score for the speech of the corresponding partial word is large, the Viterbi score for non-speech represented by noise is often small. Therefore, in the above example,
The Viterbi score for the partial word “p, a, h, u, sy, a, i, N” is smaller than the Viterbi score expected when a normal speech section is converted into a partial word sequence.

Therefore, as described above, for the partial word sequence output from the partial word sequence generation unit 12, it is registered whether or not to be registered in the user registration word dictionary 13 based on the Viterbi score of the sequence. By making a determination in the condition determination unit 21, it is possible to automatically prevent an erroneous partial word sequence having a bad score from being registered in the user registration word dictionary 13.

Various methods can be considered for the registration judgment method in the registration condition judgment unit 21, that is, the method for judging whether or not to register the partial word sequence generated by the partial word sequence generation unit 12 in the user registration word dictionary 13. Can be In the present invention, the determination method does not matter, but registration can be limited by the number of partial word sequences, for example, as described below.

In many cases, a number of word sequences having a relatively large Viterbi score of a partial word sequence appear. The partial word sequence generation unit 12 in the first embodiment outputs all partial word sequences having a Viterbi score larger than g (T).

However, in the case of a device having a strong memory (storage area) restriction, the number of partial word sequences per word is limited in order to make the area occupied by one word in the user registered word dictionary 13 as small as possible. Want to.

In such a case, the registration condition determination unit 21
The registration judgment in is as follows. Here, N is the maximum number of partial word sequences per word.

First, when the number of partial word series is N or less, all are registered in the user registration word dictionary 13. On the other hand, when the number of partial word sequences is N or more, the partial word sequences are arranged in descending order of the Viterbi score, and the upper N words are registered in the user registration word dictionary 13.

As a result, the number of partial word sequences registered in the user registered word dictionary 13 can be limited.

[Third Embodiment] As described in the second embodiment, when the data is converted into a partial word sequence including a non-speech section due to the influence of noise or the like, the recognition process is not performed. Recognition performance in the main voice recognition unit (14) is reduced.

In many cases, the partial word HMM cannot be successfully matched against noise, so that the Viterbi score of such a partial word sequence is small. However, depending on the type of noise, the partial word HMM is accidentally matched with the partial word HMM. As a result, the Viterbi score of the partial word sequence may increase. In such a case, the registration determination by the registration condition determination unit 21 applied in the second embodiment cannot prevent registration of an erroneous partial word sequence.

However, since the partial word sequence generally corresponds to the partial word sequence corresponding to the voice uttered by the speaker used, such an error can be found by the user confirming the partial word sequence.

The third embodiment implements a mechanism for preventing such erroneous registration of a partial word sequence by confirming it by a user.

FIG. 15 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a third embodiment of the present invention. In FIG. 15, the same parts as those in FIG.

The feature of the configuration of FIG. 15 is that the registration confirmation unit 31 is used instead of the registration condition determination unit 21 in the configuration of FIG. 14, that is, the registration confirmation unit 31 is added to the configuration of FIG. There is in the point. The partial word sequence output from the partial word sequence generation unit 12 is sent to the registration confirmation unit 31.

As shown in FIG. 16, the registration confirmation unit 31
Partial word sequence character string conversion unit 311, partial word sequence display character string correspondence table 312, user operation unit 313, character string display processing unit 314, display 315, and user operation determination unit 31
6 is comprised.

The partial word sequence character string conversion unit 311 converts the partial word sequence output from the partial word sequence generation unit 12 into a character string of a sequence that is easy for the user to understand. In the present embodiment, an example will be described in which a hiragana is displayed after being converted into hiragana so that the user can easily confirm it. This partial word series character string conversion unit 311
Is, as shown in FIG. 17, the partial word sequence Hiragana conversion unit 3
11a and a partial word sequence hiragana correspondence table 311b.

FIG. 18 shows an example of the partial word sequence Hiragana correspondence table 311b. In the example of FIG. 18, a partial word sequence that can be converted to hiragana and a corresponding hiragana (character code) are registered as a set in the partial word sequence hiragana correspondence table 311b.

The partial word sequence Hiragana converter 311a operates as follows according to the flowchart of FIG.

First, the hiragana conversion unit 311a inputs the partial word sequence sent from the partial word sequence generation unit 12 (step S31). X = [x (1)
, X (2), ..., x (J (X))]. A character string (result character string) obtained as a result of the conversion is defined as S.

Next, the hiragana conversion section 311a empties the conversion buffer a and the result character string buffer S, and initializes a pointer i pointing to a partial word in the partial word sequence X to 1 (steps S31, S32).

Next, the hiragana conversion unit 311a calculates i = 1 to i
= J (X) until steps S35 to S40 described below are repeated. If i exceeds J (X) (step S34),
A series of processing ends.

That is, if i is equal to or less than J (X) (step S34), the hiragana conversion unit 311a firstly sets the partial word sequence X
Is added to the conversion buffer a (step S35).

Next, the hiragana converter 311a increases i by one.

Next, the character string converter 311 searches the partial word sequence Hiragana correspondence table 311b for a partial word sequence (or partial word) equal to the partial word sequence (or partial word) in the conversion buffer a (step S37). If found (step S38), the process proceeds to step S39. On the other hand, if not found, the process returns to step S34.

In step S39, a hiragana character string corresponding to the partial word sequence in the conversion buffer a is obtained from the partial word sequence hiragana correspondence table 311b, and the resulting character string buffer S
After the contents of the buffer a are deleted (step S40), the process returns to step S34.

By repeating the above operation from i = 1 to i = J (X), the conversion of the partial word sequence X into the hiragana character string is completed, and the result of the conversion into the result character string buffer S (the hiragana character string) Is obtained.

The hiragana character string converted from the partial word sequence by the partial word sequence character string conversion unit 311 (of which the hiragana conversion unit 311a is included) is registered in the partial word sequence display character string correspondence table 312 in pairs with the partial word sequence. At the same time, it is sent to the character string display processing unit 314. FIG. 20 shows an example of registration in the partial word series display character string correspondence table 312.

The user operation unit 313 in FIG. 16 indicates a key switch for “move up the cursor”, a key switch for “move down the cursor”, and “yes” (here, “register”). A key switch has a key switch (neither is shown) that means "deny"("notregistered" here). When any key switch is pressed, the operation information is output to the character string display processing unit 314.

At the time of word registration, the character string display processing section 314 displays a word registration confirmation screen for the word specified by the user on the display 315, and displays the conversion output from the partial word series character string conversion section 311 on the screen. Character string (Hiragana character string) to be displayed.

FIG. 21 shows a display example of the word registration confirmation screen. In this word registration confirmation screen, an instruction as to whether or not to register the character string is input for each character string display field (character string display field) 211 output from the partial word sequence character string conversion unit 311. A registration instruction column 212 is provided, and another registration instruction column (decision column) 213 for determining the input content of each registration instruction column 212 is provided. These registration instruction columns 212 and 213 form a registration instruction input field 214. A cursor 215 that can move up and down in the registration instruction input field 214 is displayed on the word registration confirmation screen.

The character string display processing unit 314 receives the user's operation information from the user operation unit 313, and changes the display as follows in accordance with the operation information.

In the case of "move on cursor", the cursor 2
If the registration instruction column 212 is located above the line 15, the cursor 215 is moved to the registration instruction column 212 immediately above.

In the case of "move under cursor", the cursor 2
If there is a registration instruction column 212 or 213 below the line 15, the cursor 215 is moved to the registration instruction column 212 or 213 immediately below. If the destination is the registration instruction column 213, that is, the determination column 213, the hiragana string marked with “register” is output to the user operation determination unit 316, and the operation is terminated.

In the case of “yes”, the “register” mark (here, “register”) is registered in the registration instruction field 212 where the cursor 215 is located.
○ mark).

[0174] In the case of "No", a mark of "do not register" (here, x mark) is added to the registration instruction column 212 where the cursor 215 is located.

In the display example of FIG. 21, the user uses the word “employee”.
Request for registration, and in response to the word "employee", "shain"
Is generated, the partial word sequence generation unit 12
From "g, a, b, a, sy, a, i, N", "g, a,
d, a, sy, a, i, N "," g, a, b, a, sy,
3 shows a word registration confirmation screen when three partial word sequences “a, i, g, u” are output. Each partial word sequence is converted to a hiragana character string by a partial word sequence character string conversion unit 311 in the registration confirmation unit 31, and “gabashan”, “gadashain”, and “gabashaig” are converted. It is displayed in the character string display field 211. Here, “O” that means “register” is displayed in the registration instruction column 212 corresponding to the character string display column 211 in which “Gabashain” is displayed, and “Gadashain” is displayed. Character string display field 2
In the registration instruction column 212 corresponding to “11”, the cursor 21 is used to select and specify “register” or “not register”.
5 has been moved.

The user operation judging section 316 sets the cursor 21
As a result, the character string display processing unit 31
The character string output from No. 4, ie, the character string designated to be “registered” by the user, is converted into a partial word sequence using the partial word sequence display character string correspondence table 312 (as shown in FIG. 20). After conversion, the partial word sequence is registered in the user registration word dictionary 13.

In the display example of FIG. 21, the display character string is within the display width of the screen. However, when the display character string is longer than the display width of the screen, a groove for scrolling left and right is provided. Or, it may be folded and displayed on a plurality of lines. When the number of display character strings is larger than the number of lines on the screen, a mechanism for scrolling up and down may be provided.

As described above, in the present embodiment, since a word input by the user by voice is converted into a partial word sequence and then registered in the dictionary, the partial word sequence is known to the user before the dictionary registration. By converting the character string into an easy-to-use character string (here, a hiragana character string) and presenting it to the user, the user recognizes the content to be voice-registered, that is, the partial word sequence output from the partial word sequence generation unit 12. It is possible to check in advance whether the sequence is erroneous at the character string level (instead of the level of the partial word sequence that is difficult for the user to understand), and to check that the incorrect partial word sequence is registered. Can be blocked.

When the partial word sequence is converted to a character string and presented to the user, in addition to displaying the character string, a voice representing the character string is output by rule synthesis or the like and presented to the user. It is also possible.

[Fourth Embodiment] Next, a fourth embodiment of the present invention will be described. In the fourth embodiment, even when an erroneous partial word sequence is output, the user corrects the erroneous partial word sequence to a correct partial word sequence by a simple editing operation. This realizes a mechanism that enables registration of a correct partial word sequence.

FIG. 22 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a fourth embodiment of the present invention. The same parts as those in FIG. 15 are denoted by the same reference numerals.

The feature of the configuration of FIG. 22 is that a registration editing unit 41 is used instead of the registration confirmation unit 31 in the configuration of FIG. 15, that is, the registration editing unit 41 is added to the configuration of FIG. On the point. The partial word sequence output from the partial word sequence generation unit 12 is sent to the registration editing unit 41.

The registration / editing section 41, as shown in FIG.
(A partial word sequence character string conversion unit 311 in FIG. 17, which is a component of the registration confirmation unit 31 in the third embodiment.
(Having the same configuration as the above), a partial word sequence character string conversion unit 411, a user operation unit 413, a character string display processing unit 414, and a display unit 41
5 and a character string partial word sequence conversion unit 416.

The user operation unit 413 includes a key switch meaning “move cursor”, a key switch meaning “move cursor”, a key switch meaning “move cursor left”, and “move cursor right”. Key switch, which means "field switching",
A key switch that means "yes"("register")
Key switch for "negative"("do not register")
It has a key switch that means "delete" of a character, and a key switch (both not shown) corresponding to each "hiragana character". When any key switch is pressed, the operation information is output to the character string display processing unit 414.

The character string display processing unit 414 displays a word registration edit screen on the display 415 when confirming the word registration of the word specified by the user, and displays the part output from the partial word sequence generation unit 12 on the screen. A character string (Hiragana character string) corresponding to the word series is displayed. The character string corresponding to the partial word sequence is output from the character string conversion unit 411 by the partial word sequence character string conversion unit 411 in the same conversion operation as the partial word sequence character string conversion unit 311 having the configuration shown in FIG. Things. The conversion from the partial word sequence to the character string (Hiragana character string) in the character string conversion unit 411 is performed by the contents of the partial word sequence Hiragana correspondence table 311b in the partial word sequence character string conversion unit 311 (see FIG. 18). A partial word sequence Hiragana correspondence table (not shown) having the same contents as the above is used.

FIG. 24 shows a display example of the word registration / edit screen. In this word registration edit screen, a character string edit field 241 for displaying and editing each character string output from the partial word sequence character string conversion unit 411 and a character string on the character string edit field 241 are registered. A registration instruction input field 242 is provided for inputting an instruction as to whether or not the instruction is valid. The registration instruction input field 242 includes a registration instruction column 243 provided corresponding to each character string on the character string editing field 241, and another registration instruction column (decision) for determining the input content of each registration instruction column 243. Column) 244. In addition, a cursor 245 that can move in the character string editing field 241 and the registration instruction input field 242 is displayed on the word registration editing screen.

The character string display processing unit 414 receives user operation information from the user operation unit 413, and changes the display accordingly.

First, the operation when the cursor 245 is within the registration instruction input field 242 is as follows.

In the case of "move on cursor", the cursor 2
If the registration instruction column 243 is located above the line 45, the cursor 245 is moved to the registration instruction column 243 immediately above.

In the case of "move under cursor", the cursor 2
If there is a registration instruction column 243 or 244 below the 45 line, the cursor 245 is moved to the registration instruction column 243 or 244 immediately below. When the moving destination is the registration instruction column 244, that is, the decision column 244, the hiragana string marked with “register” is output to the character string partial word sequence conversion unit 416, and the operation is terminated.

In the case of “affirmation”, the “register” mark (here, “register”) is registered in the registration instruction field 243 where the cursor 245 is located.
○ mark).

[0192] In the case of "No", a mark of "not registered" (here, x mark) is added to the registration instruction field 243 where the cursor 245 is located.

In the case of "field switching", the cursor 245 is moved to the head position of the corresponding character string on the character string editing field 241.

In other cases, this is ignored.

On the other hand, the operation when the cursor 245 is within the character string editing field 241 is as follows.

In the case of "move cursor right", the cursor 2
If the character 45 is to the right of the character, the cursor 245 is moved one character to the right.

In the case of "move cursor left", the cursor 2
If there is a character 45 to the left of a character, the cursor 245 is moved one character to the left.

In the case of "delete", the character on which the cursor 245 is located is erased, and all the characters on the right side are shifted to the left by one character.

In the case of any of the "hiragana characters", the right is shifted one character at a time from the character where the cursor 245 is present to the right, and the hiragana character is displayed at an empty position (the position where the cursor 245 was originally).

In the case of “field switching”, the cursor 245 is moved to the corresponding registration instruction column 243 in the registration instruction input field 242.

In other cases, this is ignored.

The display example of FIG. 24 is similar to that of the third embodiment when the user requests registration of the word “employee” and utters “shain” in response to the word “employee”. , And “g, a, b,
a, sy, a, i, N "," g, a, d, a, sy, a,
9 shows a word registration / edit screen when three partial word sequences of “i, N” and “g, a, b, a, sy, a, i, g, u” are output. Each partial word sequence is stored in the partial word sequence character string conversion unit 41 in the registration / editing unit 41.
The character string is converted into a hiragana character string in step 1, and “GABA SHAIN”, “GADA SHAIN”, and “GABA SHAIGU” are displayed in the character string edit field 241. Here, “x”, which means “do not register”, is displayed in the registration instruction column 243 corresponding to the display character string “gabashan”, and “da” in the display character string “gadashain” is displayed. The cursor 245 has been moved to the position.

In this state, the user operates the user operation unit 413
Is operated and the "delete" key switch is pressed, the character string display processing unit 414 deletes "da" in the character string "gadashiin". As a result, the character string “gadashiin” becomes “gadashiin”. Further, when the user moves the cursor 245 to the position of “ga” in “gasha” and presses the “delete” key switch, the character string display processing unit 414 causes the character string “gashain” to be displayed. Delete "ga". In this way, the character string "gabashan" is edited on the character string edit field, and corrected to the correct hiragana character string "shain" for the input voice of the word "employee" as shown in FIG. Can be.

In this state, when the user presses the "field switching" key switch, the cursor 245 is moved to the registration instruction column 243 corresponding to the character string "shain". Further, when the user presses the “YES” key switch, FIG.
As shown in FIG. 5, a mark (○) of “register” is displayed in the registration instruction column 243 corresponding to the character string “shain”.

When the user moves the cursor 245 to the registration instruction column 243 corresponding to the character string “gabashiaigu” in the character string editing field 241, and presses the “negation” key switch, the registration starts. As shown in FIG. 25, a mark (×) of “do not register” is displayed in the instruction column 243.

In this state, when the user moves the cursor 245 to the decision box 244, the character string display processing unit 414 changes the hiragana string “Shain” with the mark “O” to the character string. The character string partial word sequence conversion unit 416 that is output to the partial word sequence conversion unit 416 is output from the character string display processing unit 414 by the reverse operation of the partial word sequence character string conversion unit 411 in the third embodiment. The hiragana character string is converted into a partial word sequence and registered in the user registration word dictionary 13.

Here, character string partial word series conversion section 416
Will be described in detail.

The character string partial word sequence conversion section 416 performs the processing shown in FIG.
As shown in FIG. 6, a hiragana partial word sequence conversion unit 416a,
And a partial word sequence Hiragana correspondence table 416b. The contents of the partial word sequence Hiragana correspondence table 416b are as follows.
The contents of the partial word sequence hiragana correspondence table 311b in the partial word sequence character string conversion unit 311 having the configuration shown in FIG. 18 (see FIG. 18).
Therefore, the partial word sequence hiragana correspondence table can be shared by the partial word sequence character string conversion unit 411 and the character string partial word sequence conversion unit 416.

The operation of converting a hiragana character string into a partial word sequence by the hiragana partial word sequence conversion unit 416a in the character string partial word sequence conversion unit 416 is performed as follows.

First, let the hiragana character string be S and the length be J
(S), the i-th hiragana is represented by S (i). Let X be (the buffer of) the partial word sequence obtained as a result of the conversion.

(1) X is emptied.

(2) i is incremented by 1 from 1 to J (S), and (3) is repeatedly executed.

(3) Partial word sequence Hiragana correspondence table 416b
, A partial word sequence corresponding to Hiragana S (i) is searched, and the partial word sequence is added to X.

As described above, in the present embodiment, a word input by the user by voice is converted into a partial word sequence and then registered in the dictionary. Therefore, before the dictionary is registered, the partial word sequence is transmitted to the user. The contents are converted into an easy-to-understand series of character strings (here, hiragana character strings), presented to the user, and provided to the user for editing the character strings. Sequence generation unit 1
Even if the partial word sequence output from 2 is a sequence with a recognition error, check it in advance and correct it at the character string level (instead of the partial word sequence level that is difficult for the user to understand). can do. Moreover, the corrected character string can be automatically converted into a partial word sequence and registered. Therefore, in this embodiment,
It can be said that the registered contents can be edited by character string editing.

[Fifth Embodiment] Next, a fifth embodiment of the present invention will be described.

In the embodiments described above, for example, the first embodiment, there are few problems when the user registers words as read in Chinese characters. However, in the case where registration is performed using an original reading / abbreviation, there is a possibility that the user himself / herself forgets how the registration is performed, and in that case, there is a problem. Further, when a plurality of users use one voice recognition device, there is a problem that the contents registered by other users cannot be understood. Therefore, enabling the user to confirm the registered voice content is very useful in maintaining and managing the recognition device. If there is another speech recognition device that can register characters, the confirmed (displayed) character string is re-registered in that speech recognition device, so that the registered contents by voice can be registered in another speech recognition device. You can copy easily.

In the fifth embodiment, the user registration word dictionary 1
3 realizes a mechanism that enables the registered contents to be presented to the user in an easy-to-understand form.

FIG. 27 is a block diagram showing a sub-word type speaker-independent speech recognition apparatus according to a fifth embodiment of the present invention. The same parts as those in FIG. 1 are denoted by the same reference numerals.

The configuration of FIG. 27 is characterized in that a user word registration dictionary display section 51 is added to the configuration of FIG. The user registration word dictionary 13 in FIG.
It is assumed that a registration format as shown in FIG. 9, that is, a registration format in which a pair of partial word sequences forming a word corresponding to a word name is registered.

[0220] The user word registration dictionary display section 51 converts the information of the partial word series registered in the user registration word dictionary 13 into character information of a series that is easy for the user to understand, for example, a hiragana character string and provides the user with the information. As shown in FIG. 28, as shown in FIG. 28, a partial word sequence character string conversion unit 511 (having the same configuration as the partial word sequence character string conversion unit 311 in the third embodiment), a character string display processing unit 514, And display 5
15 is comprised.

The partial word sequence character string conversion unit 511 reads a pair of a word name and a partial word sequence from the user registered word dictionary 13 and converts the partial word sequence in the same manner as the partial word sequence character string conversion unit 311. And a pair of the word name read from the user registration word dictionary 13 and a character string obtained by converting the corresponding partial word sequence is converted into a character string display processing unit 5.
14 is output.

Therefore, if the contents of the user registration word dictionary 13 are as shown in FIG. 9, the word “employee” and the character string “yain” are paired, and the word “officer” and the character string “yain” The pair of “Kin” and the pair of the word “officer” and the character string “Yain” are output to the character string display processing unit 514.

The character string display processing unit 514 displays a list of pairs of word names and character strings output from the partial word sequence character string conversion unit 511 on the display 515. This allows the user to
The registered contents of the user registered word dictionary 13 can be easily confirmed. This display example is shown in FIG.

[0224] In the fifth embodiment described above,
Although the case where the user word registration dictionary display unit 51 is added to the configuration of FIG. 1 has been described, the configuration of FIG. 2, the configuration of FIG.
It is also possible to add a user word registration dictionary display unit 51 to the configuration of FIG. 15 or the configuration of FIG.

[Sixth Embodiment] Next, a sixth embodiment of the present invention will be described.

In the fifth embodiment, by presenting the contents of the user registration word dictionary 13 in a format that is easy for the user to understand, the user can easily confirm the registration contents. However, in the fifth embodiment, when an erroneous partial word sequence is generated by the partial word sequence generation unit 12 and registered in the user registration word dictionary 13, a function of editing (changing or deleting) the partial word sequence is provided. Because it does not have
May adversely affect the recognition performance.

Therefore, the sixth embodiment realizes a mechanism that can not only confirm the registered contents of the user registered word dictionary 13 but also edit the registered contents.

FIG. 30 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a sixth embodiment of the present invention. The same parts as those in FIG. 27 are denoted by the same reference numerals.

The configuration of FIG. 30 is characterized in that a user word registration dictionary editing unit 61 is used instead of the user word registration dictionary display unit 51 in the configuration of FIG. In that a user word registration dictionary editing unit 61 is added.

The user word registration dictionary editing unit 61 reads the
As shown in the figure, a partial word sequence character string conversion unit 611 (having the same configuration as the partial word sequence character string conversion unit 311 in the third embodiment), a user operation unit 613, a character string display processing unit 614, a display 615 and a dictionary operation unit 616.

The user registration word dictionary 13 in FIG. 30 applied in the present embodiment has a registration format as shown in FIG.
That is, in addition to a registration format in which a pair of a partial word sequence forming a word corresponding to a word name is registered, a unique number (hereinafter, referred to as a word number) is assigned to each registered content as shown in FIG. Shall be attached.

The partial word sequence character string conversion unit 611 in the user word registration dictionary editing unit 61 reads a pair of a word name and a partial word sequence from the user registered word dictionary 13 having the structure shown in FIG. Performs the same conversion as that of the partial word series character string conversion unit 311 to obtain the word number assigned to the word name, the word name, and the character string obtained by converting the corresponding partial word series. Is output to the character string display processing unit 614.

By the way, the user operation section 613 is provided with a key switch meaning "move up cursor" and a "move down cursor".
, A key switch meaning “cursor left”, a key switch meaning “cursor right”, a key switch meaning “field switching”, a key switch meaning “delete item”, “ It has a key switch that means "change of item", a key switch that means "delete" of characters, and a key switch corresponding to each "hiragana character" (none is shown). When any key switch is pressed, the operation information is output to the character string display processing unit 614.

When editing the user-registered word dictionary 13, the character-string display processing unit 614 displays a user-registered word dictionary editing screen on the display 615, and outputs the partial-word-sequence character-string conversion unit 611 on the screen. List of word names and character strings to be used.

FIG. 33 shows a display example of the user registration word dictionary editing screen. The user registration word dictionary editing screen includes a word name display field 331 for displaying word names,
A character string edit field 332 for displaying and editing a character string output from the partial word sequence character string conversion unit 611 in combination with the word name on the word name display field 331.
And an edit instruction input field 333 for inputting an instruction on whether or not to edit (here, change or delete) the character string on the character string edit field 332. The edit instruction input field 333 is provided corresponding to each character string on the character string edit field 332,
An edit instruction column 334 for inputting edit instruction contents (change or deletion), and another edit instruction column (decision column) for starting a dictionary operation by the dictionary operation unit 616 according to the input contents of each edit instruction column 334. 335. Also,
On the word registration edit screen, a cursor 336 that can move in the character string edit field 332 and the edit instruction input field 333 is displayed.

The character string display processing unit 614 receives user operation information from the user operation unit 613 and changes the display accordingly.

First, the operation when the cursor 336 is within the editing instruction input field 333 is as follows.

In the case of "move on cursor", the cursor 3
If the edit instruction column 334 is located above the line at which 36 is located, the cursor 336 is moved to the edit instruction column 334 one level higher.

In the case of “move under cursor”, the cursor 3
If there is an edit instruction column 334 or 335 below the 36 line, the cursor 336 is moved to the edit instruction column 334 or 335 immediately below. When the moving destination is the edit instruction column 335, that is, the determination column 335, for all items marked with “delete” and “change”, the operation content, word number, word name, and character string represented by the mark are displayed. The set of four is output to the dictionary operation unit 616, and the operation is terminated.

In the case of "change of item", the cursor
A mark “change” (in this case, “○”) is added to a certain edit instruction column 334.

In the case of “delete item”, the cursor 336
A mark of “delete” (here, “x”) is added to a certain edit instruction column 334, and the cursor 336 is moved to the head position of the corresponding character string on the character string edit field 332.

In other cases, this is ignored.

On the other hand, the operation when the cursor 336 is within the character string editing field 332 is as follows.

In the case of “move cursor right”, the cursor 3
If there is a character 36 to the right of a character, the cursor 336 is moved one character to the right.

In the case of "cursor left movement", the cursor 3
If there is a character 36 to the left of a character, the cursor 336 is moved one character to the left.

In the case of "delete", the character on which the cursor 336 is currently located is erased, and all the characters on the right side are shifted to the left by one character.

In the case of any of the "hiragana characters", the right of the character at which the cursor 336 is now shifted by one character to the right, and the hiragana character is displayed at an empty position (the position where the cursor 336 was originally).

In the case of “field switching”, the cursor 336 is moved to the corresponding edit instruction column 334 in the edit instruction input field 333.

In all other cases, this is ignored.

The display example of FIG. 33 shows the contents of the user registration word dictionary 13 shown in FIG. 32, that is, the partial word series “y, a, i, N” of the word name “employee” of word number 1, and the word number The partial word series "y, a, k, u,
i, N ", a character string (hiragana character string) corresponding to the partial word sequence" y, a, p, u, i, N "of the word name" officer "of word number 3" Yain "" Yakuin "" This shows a user registration word dictionary editing screen when “Yain” is output from the partial word sequence character string conversion unit 611 together with the corresponding word number and word name.

In this state, the user operates the user operation unit 613.
By performing an appropriate editing operation using the key switch of, a user registration word dictionary editing screen such as that shown in FIG. 34 can be obtained.

The screen shown in FIG. 34 is obtained as follows.

First, as shown in FIG. 33, a mark (変 更) of “change” is displayed in the decision column 335 corresponding to the character string “Yain” on the first line on the character string editing field 332. Then, the cursor 336 is moved to the position of the first character “Ya” of the character string “Yain” on the first line. In this state, press the “Delete” key switch to delete “Ya”, and then use the “Hiragana” key switch to change “
By inputting “@”, the character string “yain” is corrected to “shain” as shown in FIG. Next, the cursor 336
Is moved to the decision column 335 corresponding to the character string “Yain” on the third line on the character string edit field 332, and a mark (×) of “delete” is displayed in the decision column 335.
FIG. 34 shows the user registration word dictionary editing screen at this time.

In this state, when the cursor is moved to the decision box 335, the character string display processing unit 614 sets the information of the line marked with “○”, that is, the set of “change, word number 1, employee, shashin”. And information of the line marked with “x”, that is, a set of “delete, word number 3, officer, yain” is output to the dictionary operation unit 616.

The dictionary operation section 616 is a character string display processing section 6
An information set consisting of the operation content, word number, word name, and character string is received from 14, and the user registered word dictionary 13 is operated in accordance with the information set as follows.

First, when the operation content in the received information set is “change”, the dictionary operation section 616 stores the registered content of the item having the word number in the set information in the user registration word dictionary 13 (see FIG. 32), and replaces the part of the partial word series of the registered content with the character string in the set information converted to the partial word series. Therefore, in the case where the content of the information set is “change, word number 1, employee, employee”, as is clear from FIG. 32, the partial word series “y” in the registered content of the item of word number 1 , A, i,
N "is replaced with" sy, a, i, N ". In addition,
The conversion from the character string to the partial word sequence by the dictionary operation unit 616 is performed by the character string partial word sequence conversion unit 416 (in the hiragana partial word sequence conversion unit 416 in the fourth embodiment).
It can be performed in the same manner as in a).

Next, when the operation content in the received information set is “delete”, the dictionary operation unit 616 reads the registered content of the item having the word number in the set information from the user registered word dictionary 13. Search and delete the registered contents (item) from the user registered word dictionary 13. Therefore, when the content of the information set is “deletion, word number 3, officer, yapain”, the item of word number 3 is deleted as is clear from FIG.

As a result, the registered contents of the user registered word dictionary 13 after the above-mentioned operation in the dictionary operating section 616 is changed from the state of FIG. 32 to the state of FIG.

The user word registration dictionary editing unit 61 applied in the sixth embodiment described above is similar to the configuration of FIG. 2, the configuration of FIG. 14, the configuration of FIG. 15, or the configuration of FIG. Applicable to

[Seventh Embodiment] Next, a seventh embodiment of the present invention will be described.

In the conventional sub-word type unspecified speaker's speech recognition device, the registration of a word is performed by inputting the reading in characters. In particular, words (reading for general names of operation commands) registered at the time of system design are often entered by a system designer as character strings and registered.

On the other hand, the embodiment described above, for example, the first embodiment
The user registration word dictionary 13 realized by the word registration by voice applied in the embodiment is also expressed by the partial words.

Therefore, by unifying the expression form of both word registration information, different means (voice and character) can be used.
Can be used without distinction and used for recognition. In other words, words registered at the time of system design and words registered by the user can be used without distinction and used for recognition.

The seventh embodiment makes it possible to use the words registered at the time of system design and the words registered by the user without distinguishing them, thereby simplifying the configuration of the main speech recognition unit. It is like that.

FIG. 36 is a block diagram showing a sub-word type speaker-independent speech recognition apparatus according to a seventh embodiment of the present invention. In FIG. 36, the same parts as those in FIG.

The configuration of FIG. 36 is characterized in that a character registration word dictionary 73 having the same expression format (registration format) as the user registration word dictionary 13 is added to the configuration of FIG. The main voice recognition unit 74 that recognizes input voice using both the user registered word dictionary 13 and the character registered word dictionary 73 is used instead of the main voice recognition unit 14.

The character registration word dictionary 73 is created by using a character string partial word sequence conversion unit 75, for example, as shown in FIG.

The character string partial word sequence conversion unit 75 has the same conversion function as the character string partial word sequence conversion unit 416 of the fourth embodiment shown in FIG. 26, and is input from a keyboard or the like. String (here hiragana string)
It is converted into a partial word sequence based on a partial word sequence Hiragana correspondence table (not shown) having the same contents as the partial word sequence Hiragana correspondence table 311b shown in FIG.

The character string partial word series conversion unit 75 also has a word HMM generation function similar to the word HMM generation unit 126 in the first embodiment, and has the same partial HMM generation function as that shown in FIG. A word-speech model of a word composed of an input character string by connecting (partitions of) partial words HMM registered in the partial word HMM dictionary according to a partial word sequence converted from the input character string using an HMM dictionary (Parameter of the word HMM) is generated. The character string partial word series conversion unit 75 registers the word HMM (parameter) generated in this manner in the character registration word dictionary 73 in a pair with the word name of the word formed by the input character string.

FIG. 38 shows an example of the character registration word dictionary 73. In FIG. 38, in order to register the word “external”, the user inputs a character string “squatting”,
A registration example is shown in the case where a character string “Shanai” is input for the registration of the word “sai”.

The part for creating the character registration word dictionary 73 (character string partial word series conversion unit 75) may be incorporated in the speech recognition apparatus of FIG. 36 or may not be incorporated in the speech recognition apparatus. I do not care. In the latter case, the contents of the created character registration word dictionary 73 are recorded on a removable recording medium such as a floppy disk, CD-ROM, etc., and mounted on a voice recognition device, or voice recognition is performed via a communication line or the like. What is necessary is just to load into the storage device in an apparatus.

The registration form shown in FIG. 7 is applied to the user registration word dictionary 13 in the present embodiment.
The registration format of the user registration word dictionary 13 shown in FIG. 7 is the same as the registration format of the character registration word dictionary 73 shown in FIG. 38, and a pair of a word name and a word HMM (parameter thereof) is registered. Format.

For this reason, the main speech recognition section 74 converts the character registration word dictionary 73 into the user registration word dictionary 13 in the recognition processing.
Can be used as well. Therefore, despite using both the user registered word dictionary 13 and the character registered word dictionary 73 as the main speech recognition unit 74, for example, FIG.
The same configuration as the configuration of the main voice recognition unit 14 shown in FIG. However, in the main voice recognition unit 74, (FIG.
An HMM recognizing unit (corresponding to the HMM recognizing unit 143 in FIG. 2) refers to both the user registered word dictionary 13 and the character registered word dictionary 73, and determines the Viterbi score for all the words included in both dictionaries. Is different.

Next, a specific example of recognition processing in the main speech recognition unit 74 using both the user registered word dictionary 13 and the character registered word dictionary 73 will be described.

In the recognition process, it is assumed that the user has input "Shain" by voice. In this case, for the label sequence generated from the voice uttered “Shain”, referring to the user registration word dictionary 13 having the registration contents of FIG. Word HMM (2)
38 is calculated, and the Viterbi score of the word HMM of "in-house" and the word HMM of "outside" is calculated with reference to the character registration word dictionary 73 having the registered contents of FIG. Here, the Viterbi score of "Employee" is -40,
"Executive" Viterbi scores -80 and -100, "Outside"
Has a -70 Viterbi score and "in-house" has a Viterbi score of-
If it is 75, the recognition result is the word "employee".

[0276] Next, it is assumed that the user has voice-inputted "smart". Also in this case, a Viterbi score is similarly calculated for a label sequence generated from a voice uttered “shy”. If the Viterbi score of "employee" is -90, the Viterbi score of "executive" is -75 and -70,
Assuming that the Viterbi score of “outside” is −55 and the Viterbi score of “inside” is −35, the recognition result is the word “internal”.

In the above example, the user-registered word dictionary 13 and the character-registered word dictionary 73 are completely separate, but if it is not necessary to distinguish between words registered by characters and words registered by voice, Alternatively, both may be held in the same area.

For example, the registered contents of the user registered word dictionary 13 shown in FIG. 7 and the character registered word dictionary 7 shown in FIG.
3 is stored in the area of a common dictionary (hereinafter referred to as a character / speech registered word dictionary),
The voice registration word dictionary is as shown in FIG.

Also, as shown in FIG.
An attribute indicating which means has been registered in the voice registration word dictionary, for example, a flag indicating whether it has been registered as a character or registered as a voice, is stored for each registered content, so that it is possible to distinguish between the two. Even if there is, both can be mixed and held in the same area. Note that FIG.
In the example of (b), only the registration means is represented by a flag, but information on the registered date and time can be registered as an attribute.

The configuration in which the user registered word dictionary 13 and the character registered word dictionary 73 applied in the seventh embodiment described above are used together is the configuration shown in FIG. 2, the configuration shown in FIG. 14, the configuration shown in FIG. The same applies to the configuration of FIG. 22, the configuration of FIG. 27, or the configuration of FIG.

[Eighth Embodiment] Next, an eighth embodiment of the present invention will be described.

In the above-described embodiment, when the user makes a special utterance (for example, a strong rounding), the main voice recognition unit (1)
The recognition accuracy of 4) may decrease. In such a case, the likelihood (Viterbi score) of each word obtained by the recognition processing in the main speech recognition unit (14) tends to decrease.
If there is such a tendency, the partial word sequence generation unit (1
If the recognition result of 2) is registered in the user registration word dictionary (13) and the recognition is performed using the same from the next time, the recognition accuracy of the main voice recognition unit (14) can be improved.

In the eighth embodiment, the input speech is recognized not only by the main speech recognition unit but also by the partial word sequence generation unit, and based on the result, whether or not the partial word sequence can be registered is determined. By automatically registering in the user registration word dictionary, the recognition accuracy of the main voice recognition unit can be improved.

FIG. 40 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to an eighth embodiment of the present invention. The same parts as those in FIG. 1 are denoted by the same reference numerals.

The features of the configuration shown in FIG. 40 are that in the voice recognition mode, the input voice is input not only to the main voice recognition unit 14 but also to the partial word sequence generation unit 12; A point-of-use word registration determining unit 81 that determines whether or not a partial word sequence can be registered based on the results of the recognition processing of both units and registers the partial word sequence in the user registered word dictionary 13 is newly provided. .

Also, the main speech recognition section 14 in the present embodiment
Output the recognition result and the likelihood (Viterbi score) from the previous embodiments. The partial word sequence generation unit 12 outputs the likelihood (Viterbi score) of the partial word sequence in addition to the partial word sequence, as in the second embodiment.

In FIG. 40, a mode switching unit corresponding to the mode switching unit 11 in FIG. 1 is omitted. In the word registration mode, the mode switching unit inputs the input speech to the partial word sequence generation unit 12 in the same manner as the mode switching unit 11 in FIG. 1, whereas in the speech recognition mode, the input speech is input to the main speech recognition unit. 14 and the partial word sequence generation unit 12.

In the configuration shown in FIG. 40, the input voice is input to both main voice recognition unit 14 and partial word sequence generation unit 12. The main speech recognition unit 14 performs recognition processing on the input speech using the user registered word dictionary 13 in the same manner as in the first embodiment, and outputs a recognition result and its likelihood (Viterbi score). On the other hand, as in the case of the word registration mode in the first embodiment, the partial word sequence generation unit 12 converts the input speech into a partial word sequence, and converts the partial word sequence and its likelihood (Viterbi score). Output. Here, it is assumed that words have been registered in the user registration word dictionary 13 by word registration processing in the word registration mode.

The in-use word registration determination unit 81 compares the likelihood of the recognition result output from the main speech recognition unit 14 with the likelihood of the partial word sequence output from the partial word sequence generation unit 12, and determines the latter. Is larger and the difference is larger than a predetermined reference value (threshold) Z, the partial word sequence is regarded as a partial word sequence corresponding to the recognition result of the main speech recognition unit 14, and the user registration word Register in the dictionary 13.

The operation of the in-use word registration judging section 81 will be described in detail by taking as an example a case where the contents of the user registration word dictionary 13 are as shown in FIG.

When user A tries to enter "employee",
As a result of uttering “Shain”, the output of the main voice recognition unit 14 is a word “employee” whose Viterbi score is −25, and the output of the partial word sequence generation unit 12 is a partial word sequence whose Viterbi score is −20. sy, a, i "and the Viterbi score is -25.
Is a partial word sequence “sy, a, i, N”.

The in-use word registration judgment unit 81 firstly obtains the Viterbi score -25 of the word "employee" and the partial word series "s
y, a, i "with the Viterbi score-20. Since the Viterbi score of the partial word sequence is larger, the in-use word registration determination unit 81 obtains the difference and compares it with the reference value Z.
Here, assuming that the reference value Z is set to 20, the difference 5 is smaller than Z.
No. 1 does not register “sy, a, i” in the user registration word dictionary 13.

Next, the in-use word registration judging section 81 calculates the Viterbi score -25 of the word "employee" and the partial word sequence "s
y, a, i, N "with the Viterbi score-25.
No registration is performed because the Viterbi score of the partial word sequence is not larger.

That is, "Shain" uttered by user A
Is very close to the voice expected from the partial word sequence “sy, a, i, N” for the word “employee” that was originally registered. This is because the partial word sequence generation unit 12
Contains "sy, a, i, N" in the output of the, the Viterbi score of the optimal partial word sequence "sy, a, i",
This can be determined from the fact that the Viterbi scores of “sy, a, i, N” are relatively close values. Therefore, in this case, it is not necessary to add a new item to the user registration word dictionary 13.

Next, another user B tries to input "employee" and utters "shain", and the output of the main voice recognition unit 14 indicates that the word "employee" has a Viterbi score of -55 and a partial When the output of the word sequence generation unit 12 has a Viterbi score of -20
Is a partial word sequence "sy, e, i, N" and a partial word sequence "j, e, i, N" with a Viterbi score of -45.

The in-use word registration determining unit 81 firstly obtains the Viterbi score −55 of the word “employee” and the partial word series “s
y, e, i, N "with the Viterbi score-20.
Since the Viterbi score of the partial word series is larger, the in-use word registration determination unit 81 obtains the difference and obtains the reference value Z (= 2
0). Since the difference 35 is larger than Z, the in-use word registration determination unit 81 newly registers “sy, e, i, N” in the user registration word dictionary 13 as a partial word sequence corresponding to the word “employee”.

Next, the in-use word registration determination section 81 determines the Viterbi score -55 of the word "employee" and the partial word sequence "j,
e, i, N "with a Viterbi score of -45. Since the Viterbi score of the partial word sequence is larger, the difference is obtained and compared with a constant Z (= 20). Since the difference 10 is smaller than Z, “j, e, i, N” is not registered.

In other words, "Shain" uttered by user B
Is different from the voice expected from the partial word sequence “sy, a, i, N” for the word “employee” originally registered. This is the partial word sequence generation unit 1
2 outputs the optimal partial word sequence "sy, e, i, N"
Can be determined in this way because the Viterbi score of “sy, a, i, N” greatly exceeds the Viterbi score of “sy, a, i, N”. Therefore, in this case, a new partial word sequence “sy, e,
It is appropriate to additionally register “i, N”.

As a result, the user registration word dictionary 13 having the contents shown in FIG. 41A is as shown in FIG. 41B. In the user registration word dictionary 13 shown in FIG. 41B, a new item is added to the word “employee” according to the utterance tendency of the user B.

As described above, the speech recognition apparatus of this embodiment capable of automatically registering the recognition result in the user registration word dictionary 13 functions extremely effectively when the word uttered by the user is known.

Therefore, a modified example of the configuration of the speech recognition apparatus of FIG. 40 in the case where the word uttered by the user is known will be described with reference to the same FIG. 40 for convenience. Here, we prepare a new mode called adaptive mode,
In the adaptive mode, a user interface (not shown) for presenting a word to a user and instructing the user to utter the word is provided.

The user utters words presented from the device (the user interface therein) in the adaptive mode.

The voice uttered by the user in the adaptive mode is input to both the main voice recognition unit 14 and the partial word sequence generation unit 12. The main voice recognition unit 14 uses the user registered word dictionary 13 to obtain and output a likelihood (Viterbi score) for the word that the device (user interface therein) has instructed to utter. On the other hand, the partial word sequence generation unit 12 converts the input speech into a partial word sequence, and outputs the partial word sequence and its likelihood (Viterbi score).

The in-use word registration determination unit 81 determines the likelihood of the recognition result of the main speech recognition unit 14, that is, the likelihood of the word instructed to utter, and the partial word sequence output from the partial word sequence generation unit 12. If the latter is larger and the difference is larger than the reference value Z, the partial word sequence is paired with the word instructed to be uttered and registered in the user registration word dictionary 13. I do.

Next, another modified example in which the word uttered by the user is known will be described with reference to FIG. 40 for convenience.

The voice uttered in the adaptive mode is input to both the main voice recognition unit 14 and the partial word sequence generation unit 12. The main voice recognition unit 14 is different from the above-described modified example in that the user registration word dictionary 13
Perform normal recognition using. The partial word sequence generation unit 12 converts the input speech into a partial word sequence and outputs it. Here, there is no need to output the likelihood of the partial word sequence.

The in-use word registration judging section 81 judges whether or not the recognition result of the main speech recognizing section 14 is the same as the word instructing the utterance. Is registered in the user registration word dictionary 13 by pairing the partial word sequence output from the word with the word instructed to utter.
Note that the partial word sequence generation unit 12 outputs the likelihood together with the partial word sequence, and the likelihood of the partial word sequence is added to the determination condition of the in-use word registration determination unit 81 ( That is, a combination with the function of the registration condition determination unit 21 in the second embodiment is also possible.

In the above description, in the speech recognition apparatus (in-use word registration / judgment unit 81) of FIG. 40, a new partial word sequence is generated based on both outputs of the main speech recognition unit 14 and the partial word sequence generation unit 12. Has been described, but the present invention is not limited to this. For example, a determination is made once only based on the likelihood of the recognition result of the main speech recognition unit 14, and a recognition process is performed by the partial word sequence generation unit 12 according to the determination result. By comparing with the likelihood of the partial word sequence, whether or not to register a new partial word sequence may be determined. A modification of the configuration in FIG. 40 will be described with reference to the block configuration diagram in FIG.

In the speech recognition apparatus having the configuration shown in FIG. 42, first, an input speech is input to the main speech recognition unit 14, and the recognition result and likelihood (Viterbi score) for the speech are calculated. At the same time, the input voice is temporarily stored in the input voice buffer 83.

The in-use word registration determining unit 82 (corresponding to the in-use word registration determining unit 81 in FIG. 40) determines in advance the likelihood (Viterbi score) of the recognition result of the main speech recognition unit 14. If the former is larger than the reference value Z1, it is determined that registration of a new partial word series is not performed.

On the other hand, when the latter is larger,
The in-use word registration determination unit 82 controls the input voice buffer 83 to cause the partial word sequence generation unit 12 to output the input voice temporarily stored in the buffer 83. Thereby, the partial word sequence generation unit 12 converts the input speech into a partial word sequence, and the partial word sequence and likelihood (Viterbi score)
Is output. The subsequent operation of the in-use word registration determination unit 82 is the same as that of the in-use word registration determination unit 81 described above.

That is, the in-use word registration determination unit 82 compares the likelihood of the recognition result of the main speech recognition unit 14 output earlier and the likelihood of the partial word sequence output from the current partial word sequence generation unit 12. When the latter is larger and the difference is larger than the reference value Z, the partial word sequence is regarded as a partial word sequence corresponding to the recognition result of the main speech recognition unit 14 and is registered in the user registration word dictionary 13. sign up.

In the speech recognition apparatus having the configuration shown in FIG. 42, if Z1 = -40, for example, the above-mentioned speaker A will
, The partial word sequence generation unit 12 does not operate for the voice “Shain”. On the other hand, for the voice “Shain” uttered by the speaker B, the partial word sequence generation unit 1
2 operates, and a new word is additionally registered in the user registration word dictionary 13.

With such a configuration, it is possible to eliminate the processing of the partial word sequence generation unit 12 for the average utterance, so that the main function of the speech recognition apparatus (the partial word sequence generation unit If the functions of the main voice recognition unit 12 and the main voice recognition unit 14 are realized by a computer, the load on the computer is reduced. Therefore, it is suitable for realizing the main functions of the speech recognition apparatus on a computer that is not high-speed.

The user registration word dictionary 13 in the configuration shown in FIGS. 40 and 42 includes, in addition to the one generated by setting the word registration mode and registering words in the same speech recognition device, the first dictionary described above. It is possible to use a word generated in another device by the word registration method applied in any of the seventh to seventh embodiments. In this case, the speech recognition apparatus having the configuration shown in FIGS. 40 and 42 does not necessarily need the word registration mode and the speech recognition mode. In a normal use as the speech recognition apparatus, the main speech recognition unit 14 and the partial word sequence generation The word can be automatically registered in the user registered word dictionary 13 by using the unit 12 together.

In the speech recognition apparatus according to the present embodiment, the character registration word dictionary (7
It is also possible to use 3) together with the user registration word dictionary 13. In this case, as in the seventh embodiment, the main voice recognition unit 14 performs recognition using both the character registration dictionary (73) and the user registration word dictionary 13. This main voice recognition unit 14
Depending on the likelihood of the recognition result in the partial word sequence generation unit 1
The partial word sequence from 2 is registered in the user registration word dictionary 13 by the in-use word registration determination unit 81.

[Ninth Embodiment] Next, a ninth embodiment of the present invention will be described.

As described in the eighth embodiment, when the user makes a special utterance (for example, strong rounding), the recognition accuracy of the main voice recognition unit (14) may decrease.
In such a case, the likelihood of the main speech recognition unit (14) tends to decrease. If the output of the main speech recognition unit (14) is wrong, the partial word sequence is registered in the user registration word dictionary (13) according to the user's instruction, and if the recognition is performed using it again from the next time, As in the eighth embodiment, the recognition accuracy of the main voice recognition unit (14) can be improved.

In the ninth embodiment, the input speech is recognized not only by the main speech recognition unit but also by the partial word sequence generation unit, and the partial word sequence output from the partial word sequence generation unit is registered. By making it possible for the user to indicate whether or not the recognition is performed based on the recognition result from the main voice recognition unit, the recognition accuracy of the main voice recognition unit can be improved.

FIG. 43 is a block diagram showing a subword-type speaker-independent speech recognition apparatus according to a ninth embodiment of the present invention. The same parts as those in FIG. 40 are denoted by the same reference numerals.

A feature of the configuration of FIG. 43 is that the in-use word registration confirmation unit 91 is used instead of the in-use word registration determination unit 81 in the configuration of FIG.

In the configuration of FIG. 43, the input voice is input to both the main voice recognition unit 14 and the partial word sequence generation unit 12. The main voice recognition unit 14 performs a recognition process on the input voice using the user registered word dictionary 13 in the same manner as in the first embodiment, and outputs a recognition result. On the other hand, as in the case of the word registration mode in the first embodiment, the partial word sequence generation unit 12 converts an input voice into a partial word sequence and outputs the partial word sequence. Here, it is assumed that words have been registered in the user registration word dictionary 13 by word registration processing in the word registration mode.

The in-use word registration confirmation section 91 has an input section (user operation section) which can be operated by the user, and indicates whether or not the partial word sequence is registered in the user registration word dictionary 13. From the input unit. Then, the in-use word registration confirmation unit 91 registers the partial word sequence output from the partial word sequence generation unit 12 in the user registration word dictionary 13 as a partial word sequence corresponding to the recognition result of the main speech recognition unit 14.

The details of the operation of the in-use word registration confirmation section 91 will be described in the case where the contents of the user registration word dictionary 13 are as shown in FIG. 41A as in the eighth embodiment. Will be described.

When the user tries to enter "employee",
As a result of uttering “Shain”, the output of the main voice recognition unit 14 is the word “employee”, and the output of the partial word sequence generation unit 12 is the partial word sequence “sy, e, i, N”. I do.

If the user normally uses the voice recognition device shown in FIG. 43 and feels that it is difficult to recognize the word “employee”, the user operates the input unit of the in-use word registration confirmation unit 91 to operate. , To register the partial word sequence in the user registration word dictionary 13.

Then, the in-use word registration confirmation unit 91 outputs the word “employee” output from the main speech recognition unit 14 and the partial word sequence “sy, e, i,
The pair of “N” is additionally registered in the user registration word dictionary 13.
As a result of the additional registration, the user registration word dictionary 13 having the content of FIG. 41A is as shown in FIG. 41B.

As described above, in this embodiment, when the in-use word registration determining unit 81 receives the instruction to register the partial word sequence, the partial word sequence is registered in association with the recognition result of the main speech recognition unit 14. You. However, in this method, if the recognition result of the main speech recognition unit 14 is wrong, an incorrect combination of a word name and a partial word sequence is registered in the user registration word dictionary 13.

Therefore, in order to solve such a problem, the in-use word registration determining unit 81 in FIG. 43 uses not only a registration instruction of a partial word sequence but also information on which word the partial word sequence is to be associated with. May be received. A modification of the ninth embodiment will be described by taking as an example a case where the contents of the user registration word dictionary 13 are as shown in FIG.

When the user tries to enter "employee",
As a result of uttering “Shain”, the output of the main voice recognition unit 14 is the word “employee”, and the output of the partial word sequence generation unit 12 is the partial word sequence “sy, e, i, N”. I do.

The user operates the input unit of the in-use word registration determining unit 81 to instruct registration of a partial word string,
Enter that the current utterance was for the word "employee".

When the use word registration confirmation unit 91 receives the instruction to register the partial word string and the information on the word “employee”, the received word “employee”, that is, the word “employee” specified by the user, is used. Then, a pair of the partial word sequence “sy, e, i, N” output from the partial word sequence generation unit 12 is additionally registered in the user registration word dictionary 13.

Thus, even when the recognition result of the main voice recognition unit 14 is incorrect, a correct combination of a word name and a partial word sequence is additionally registered in the user registration word dictionary 13. User registration word dictionary 13 immediately after this additional registration
Is as shown in FIG. 41 (b).

In addition to the configuration of FIG. 43, similar to the configuration of FIG. 42 in the eighth embodiment, an input speech buffer is provided on the input side of the partial word sequence The input speech may be processed by the partial word sequence generation unit 12 only when 81 receives the instruction to register the partial word sequence. This makes it possible to reduce the load on the computer when the main function of the speech recognition device is realized by a computer, as described in the eighth embodiment.

The main functions of the speech recognition apparatus applied in the above-described embodiment, such as the partial word sequence generation processing by the partial word sequence generation unit 12 and the recognition processing by the main speech recognition unit 14, are program-readable. C that records a program for causing a computer to execute the process.
It is also realized by mounting a recording medium such as a D-ROM, a floppy disk, and a memory card, and reading and executing a program recorded on the recording medium by a computer. Here, since the recognition processing by the main voice recognition unit 14 can be performed using existing voice recognition software, the program recorded on the recording medium is a program that causes the computer to execute the processing excluding the above recognition processing. Program may be used. The content of the recording medium on which the program is recorded may be downloaded to a computer via a communication line or the like.

[0336]

As described above in detail, according to the present invention, it is possible to register words that can be used by an unspecified speaker with the same ease as that of the specific speaker voice recognition system.

According to the present invention, the word dictionary can be updated in accordance with the pronunciation of the user. According to the present invention, it is determined whether or not the word sequence to be registered satisfies the registration condition. By doing so, it is possible to prevent erroneous registration from being performed in the registration of a word dictionary according to the pronunciation of the user, thereby preventing a reduction in recognition performance.

Further, according to the present invention, since the user can confirm the presence or absence of a recognition error when the partial word sequence to be registered is generated, the user can register the partial word sequence. Can be prevented.

According to the present invention, the user can confirm an error in a partial word sequence to be registered, correct the partial word sequence, and register the partial word sequence. can do.

According to the present invention, the contents registered in the user-registered word dictionary are converted into character information and presented to the user, so that the user can check the information registered in the user-registered word dictionary at a later date. be able to.

According to the present invention, in addition to converting the contents of the user registration word dictionary into character information and presenting it to the user, the presented contents are used for the user's editing operation, and the editing process is performed. By reflecting the result in the user registration word dictionary, when the user confirms that there is a problem, the problem can be corrected.

Further, according to the present invention, the expression form of the registered information in the character registered word dictionary in which information corresponding to the partial word sequence generated from the character string information representing the reading of the word is registered is changed to the user registered word dictionary. In this case, the user registration word dictionary and the character registration word dictionary are registered in different ways, respectively, by using the registration information expression form in Recognition can be performed, and the configuration of the device can be simplified.

According to the present invention, not only in the word registration mode but also in the speech recognition mode, a partial word sequence is generated for the input speech, and the partial word sequence is generated according to the recognition status of the user's utterance. By performing the additional registration, the recognition accuracy can be improved by sequentially adapting to the user's voice.

[Brief description of the drawings]

FIG. 1 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a partial word sequence generation unit 12 in FIG.

FIG. 3 is a view showing an example of a partial word connection table 123 in FIG. 2;

FIG. 4 is a view showing a part of a flowchart for explaining the operation of the partial word HMM recognition unit 124 in FIG. 2;

FIG. 5 is a diagram showing another part of the flowchart for explaining the operation of the partial word HMM recognition unit 124 in FIG. 2;

FIG. 6 is a view showing the rest of the flowchart for explaining the operation of the partial word HMM recognizing unit 124 in FIG. 2;

FIG. 7 is a view showing an example of a user registration word dictionary 13 in FIG. 1;

FIG. 8 is a block diagram showing a configuration of a main speech recognition unit 14 in FIG. 1;

FIG. 9 is a view showing another example of the user registered word dictionary 13 in FIG. 1;

FIG. 10 is a block diagram showing a configuration of a partial word sequence generation unit 12 when a user registration word dictionary 13 in the format of FIG. 9 is used.

11 is a block diagram showing a configuration of a main speech recognition unit 14 when a user registration word dictionary 13 in the format of FIG. 9 is used.

FIG. 12 is a block diagram showing a modification of the configuration of FIG. 1 when the system of partial words used by the main speech recognition unit 14 and the system of partial words used by the partial word sequence generation unit 12 are different.

FIG. 13 is a view showing an example of a partial word HMM dictionary 15 in FIG. 12;

FIG. 14 is a block diagram of a subword-type speaker-independent speech recognition apparatus according to a second embodiment of the present invention.

FIG. 15 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a third embodiment of the present invention.

16 is a block diagram showing a configuration of a registration confirmation unit 31 in FIG.

17 is a partial word sequence character string conversion unit 311 in FIG.
FIG. 2 is a block diagram showing the configuration of FIG.

18 is a partial word sequence Hiragana correspondence table 311 in FIG.
The figure which shows an example of b.

19 is a partial word sequence hiragana conversion unit 311 in FIG.
7 is a flowchart for explaining the operation of FIG.

20 is a partial word series display character string correspondence table 3 in FIG.
The figure which shows the example of registration in 12.

FIG. 21 is a diagram showing an example of a word registration confirmation screen displayed by the character string display processing unit 314 in FIG.

FIG. 22 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a fourth embodiment of the present invention.

FIG. 23 is a block diagram showing a configuration of a registration editing unit 41 in FIG. 22;

24 is a view showing an example of a word registration edit screen displayed by the character string display processing unit 414 in FIG. 23.

FIG. 25 is a diagram showing an example of a state after a character string editing process on the word registration editing screen of FIG. 24;

26 is a character string partial word sequence conversion unit 416 in FIG.
FIG. 2 is a block diagram showing the configuration of FIG.

FIG. 27 is a block diagram showing a subword-type speaker-independent speech recognition apparatus according to a fifth embodiment of the present invention.

FIG. 28 is a block diagram showing a configuration of a user word registration dictionary display unit 51 in FIG. 27;

FIG. 29 is a view showing an example of user registered word dictionary contents displayed by a character string display processing unit 514 in FIG. 28;

FIG. 30 is a block diagram of a sub-word type speaker-independent speech recognition apparatus according to a sixth embodiment of the present invention.

FIG. 31 is a block diagram showing a configuration of a user word registration dictionary editing unit 61 in FIG. 30;

FIG. 32 is a user registered word dictionary 13 to which word numbers are assigned.
The figure which shows the registration format of.

FIG. 33 is a view showing an example of a user registration word dictionary editing screen displayed by the character string display processing unit 614 in FIG. 31.

FIG. 34 is a diagram showing an example of a state after a character string editing process on the user registration word dictionary editing screen of FIG. 33;

FIG. 35 is a diagram showing an example of the contents of the user registration word dictionary 13 changed from the state of FIG. 32 by a dictionary operation according to the result of the character string editing process on the user registration word dictionary editing screen of FIG.

FIG. 36 is a block diagram showing a sub-word type speaker-independent speech recognition apparatus according to a seventh embodiment of the present invention.

FIG. 37 is a view for explaining a method of creating the character registration word dictionary 73 in FIG. 36;

FIG. 38 is a view showing an example of a character registration word dictionary 73 in FIG. 36;

FIG. 39 is a diagram showing a character / word in which the contents of the user registered word dictionary 13 and the character registered word dictionary 73 in FIG. 36 are stored in a common area;
The figure which shows an example of a voice registration word dictionary.

FIG. 40 is a block diagram showing a subword-type speaker-independent speech recognition apparatus according to an eighth embodiment of the present invention.

FIG. 41 is a diagram showing an example of the contents of a user registration word dictionary 13 before and after word registration by the in-use word registration determination unit 81 in the configuration of FIG. 40;

FIG. 42 is a block diagram showing a modification of the configuration in FIG. 40;

FIG. 43 is a block diagram showing a subword-type speaker-independent speech recognition apparatus according to a ninth embodiment of the present invention.

FIG. 44 is a block diagram of a conventional sub-word type unspecified speaker speech recognition apparatus.

FIG. 45 is a diagram showing a three-state two-loop discrete HMM.

FIG. 46 is a diagram showing an example of a storage format of a partial word HMM registered in the partial word HMM dictionary 460 in FIG.

FIG. 47 is a partial word sequence “o,
The figure which shows the word HMM corresponding to "t, o, n, a".

FIG. 48 is a diagram showing an example of a storage format of a word HMM registered in the word HMM dictionary 450 in FIG. 44.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 11 ... Mode switching part 12 ... Partial word series generation part 13 ... User registration word dictionary 14, 74 ... Main voice recognition part (word voice model acquisition means) 15, 125, 145 ... Partial word HMM dictionary 16 ... Word HMM generation part (Partial word system conversion means) 21: registration condition determination unit 31: registration confirmation unit 41: registration editing unit 51: user word registration dictionary display unit 61: user word registration dictionary editing unit 73: character registration word dictionary 81, 82 ... In-use word registration determination section 83... Input voice buffer 91... In-use word registration confirmation section 121, 141. Acoustic analysis section 122, 142. Quantization section 123. ... word HMM generation unit 143 ... HMM recognition unit 311,411,511,611 ... partial word sequence character string conversion unit

Claims (25)

[Claims] 1. A partial word sequence generating means for converting an input speech into a sequence of at least one partial word, and information for registering information corresponding to the partial word sequence converted by the partial word sequence generating means. User registered word dictionary; word speech model acquisition means for acquiring a word speech model in which partial word speech models are joined from information corresponding to each partial word sequence registered in the user registered word dictionary; A sub-word type unspecified speaker voice recognition device, comprising: main voice recognition means for recognizing the uttered voice using a word voice model acquired from the user registered word dictionary. 2. A partial word sequence generating means for converting input speech into a sequence of at least one partial word, and the partial word sequence converted by said partial word sequence generating means satisfies a predetermined registration condition. Registration condition determination means for determining whether or not the user condition is satisfied; a user registration word dictionary in which information corresponding to the partial word sequence determined to satisfy the registration condition by the registration condition determination means is registered; Word speech model acquisition means for acquiring a word speech model in which partial word speech models are joined from information corresponding to each partial word sequence registered in the registered word dictionary; and Main speech recognition means for recognizing by using a word speech model acquired from a word dictionary. . 3. A partial word sequence generating means for converting input speech into at least one partial word sequence, and for all partial word sequences converted by the partial word sequence generating means, the partial word sequences are Registration confirmation means for presenting information to the user, accepting designation from the user as to whether or not the registration is possible, and confirming whether or not registration of a corresponding partial word sequence is possible according to the accepted designation content; A user-registered word dictionary in which information corresponding to the partial word sequence whose registration has been confirmed by the confirmation unit is registered; and a partial-word speech from information corresponding to each partial word sequence registered in the user-registered word dictionary. A word-speech model obtaining means for obtaining a word-speech model in which the models are joined; and a voice uttered by the user is obtained from the user registration word dictionary. Word-type speaker-independent speech recognition apparatus characterized by comprising a recognizing main speech recognition means with the word speech model. 4. A partial word sequence generating means for converting input speech into at least one partial word sequence, and for all partial word sequences converted by said partial word sequence generating means, the partial word sequence Presenting the information to the user, accepting a user's editing operation for the information, performing an editing process on the information, reflecting the result of the editing process on the corresponding partial word sequence, and A registration / editing unit that accepts designation from the user as to whether or not registration is possible, and confirms whether or not registration of a corresponding partial word sequence is possible in accordance with the accepted designation content; A user-registered word dictionary in which information corresponding to the partial word series is registered, and corresponding to each of the partial word series registered in the user-registered word dictionary Word speech model acquisition means for acquiring a word speech model in which partial word speech models are joined from information; and a main unit for recognizing speech uttered by a user using the word speech model acquired from the user registration word dictionary. A subword type speaker-independent speech recognition apparatus, comprising: speech recognition means. 5. A user word registration dictionary display means for converting information corresponding to a partial word sequence registered in the user registration word dictionary into character information and presenting it to a user. The subword-type unspecified speaker speech recognition device according to any one of claims 1 to 4. 6. Converting information corresponding to a partial word sequence registered in the user registration word dictionary into character information and presenting it to a user, and accepting a user's editing operation on the information. 5. The apparatus according to claim 1, further comprising a user word registration dictionary editing unit that performs an editing process on the information and reflects a result of the editing process on the user registration word dictionary. 3. A subword-type speaker-independent speech recognition apparatus according to claim 1. 7. A character registered word dictionary in which information corresponding to a partial word sequence generated from character string information representing word reading is registered in the same expression form as the user registered word dictionary. The word speech model acquiring means acquires a word speech model in which partial word speech models are joined from information corresponding to each partial word sequence registered in the user registered word dictionary, and stores the word speech model in the character registered word dictionary. A word speech model in which partial word speech models are joined is acquired from information corresponding to each registered partial word sequence, and the main speech recognition unit acquires from the user registered word dictionary and the character registered word dictionary. The subword type feature according to any one of claims 1 to 4, wherein the speech uttered by the user is recognized using the word speech model obtained. Speaker voice recognition device. 8. A partial word sequence generating means for recognizing a voice uttered by a user in a word registration mode and a voice recognition mode to generate at least one partial word sequence, and generating the partial word sequence by the partial word sequence generating means. A user registered word dictionary in which information corresponding to the registered partial word sequence is registered, and a word voice obtained by joining partial word voice models from information corresponding to each partial word sequence registered in the user registered word dictionary. Word speech model acquisition means for acquiring a model; main speech recognition means for recognizing speech uttered by a user in the speech recognition mode using a word speech model acquired from the user registered word dictionary; speech recognition mode Sometimes, the likelihood of the partial word sequence generated by the partial word sequence generation unit, the recognition result of the main speech recognition unit, and the likelihood of the recognition result are small. Based on at least one, it is determined whether or not the partial word sequence generated by the partial word sequence generating means can be registered, and information on the partial word sequence is stored in the user registered word dictionary in accordance with the determination result. A subword-type unspecified speaker speech recognition apparatus, comprising a use-time word registration determination unit for additionally registering. 9. A partial word sequence generating means for recognizing a voice uttered by a user in a word registration mode and a voice recognition mode to generate at least one partial word sequence; A user registered word dictionary in which information corresponding to the registered partial word sequence is registered, and a word voice obtained by joining partial word voice models from information corresponding to each partial word sequence registered in the user registered word dictionary. Word speech model acquisition means for acquiring a model; and main speech recognition for recognizing speech uttered by a user in the speech recognition mode using a word speech model acquired from the user registered word dictionary and outputting a recognition result. Means for receiving a partial word sequence registration instruction from a user at the time of output of a recognition result by the main voice recognition means, and receiving the registration instruction. A use word registration confirming means for additionally registering information corresponding to the partial word series generated by the partial word series generating means and registered and instructed in the user registered word dictionary. Unspecified speaker voice recognition device. 10. A partial word system conversion unit for converting a partial word sequence converted by the partial word sequence generation unit into another partial word sequence having an expression system different from the partial word sequence, further comprising: 5. The sub-word according to claim 1, wherein information corresponding to the partial word sequence converted by the system conversion unit is registered in the user registration word dictionary. Type-independent speaker speech recognition device. 11. The registration condition determining means determines whether the partial word sequence satisfies a registration condition by determining the likelihood of the partial word sequence and the number of partial word sequences converted by the partial word sequence generation means. 3. The apparatus according to claim 2, wherein the determination is made based on at least one of the following. 12. A partial word sequence generation for generating a sequence of at least one partial word by recognizing a voice uttered by a user in a word registration mode and when a predetermined condition is satisfied in a voice recognition mode. Means, a user registered word dictionary in which information corresponding to the partial word sequence generated by the partial word sequence generating means is registered, and information corresponding to each partial word sequence registered in the user registered word dictionary A word-speech model obtaining means for obtaining a word-speech model in which partial word-speech models are joined from each other; Main voice recognition means for recognizing; input voice storage means for temporarily storing voice uttered by a user in a voice recognition mode; The presence or absence of the condition is determined based on the likelihood of the recognition result of the voice recognition unit.If the condition is determined to be satisfied, the voice stored in the input voice storage unit is transmitted to the partial word sequence generation unit. The partial word sequence generation unit is operated by inputting the at least one of the likelihood of the partial word sequence generated by the partial word sequence generation unit, the recognition result of the main voice recognition unit, and the likelihood of the recognition result. Is determined on the basis of the partial word sequence generation unit, whether or not the partial word sequence generated by the partial word sequence generation unit can be registered is determined, and information on the partial word sequence is additionally registered in the user registered word dictionary according to the determination result. A sub-word type unspecified speaker speech recognition apparatus, comprising: a use-time word registration determination unit. 13. A partial word sequence generating means for converting an input speech into a sequence of at least one partial word, and information corresponding to the converted partial word sequence is converted from information corresponding to the partial word sequence. Dictionary registration means for registering in a user registration word dictionary referred to at the time of speech recognition so as to acquire a word speech model in which partial word speech models are joined and perform recognition processing using the word speech model. A word dictionary creating apparatus for subword-type speaker-independent speech recognition. 14. A voice uttered by a user for word registration is converted into at least one partial word sequence, and information corresponding to the converted partial word sequence is converted to information corresponding to the partial word sequence. A sub-word registered in a user-registered word dictionary referred to during speech recognition so that a word-speech model in which partial word-speech models are connected to each other is obtained and recognition processing can be performed using the word-speech model. A method for creating a word dictionary for type-independent speaker speech recognition. 15. A voice uttered by a user for word registration is converted into a sequence of at least one partial word, and it is determined whether or not the converted partial word sequence satisfies a predetermined registration condition. Determining the information corresponding to the partial word sequence determined to satisfy the registration condition, obtaining the word voice model in which the partial word voice models are joined from the information corresponding to the partial word sequence, and A method of creating a word dictionary for subword-type unspecified speaker speech recognition, characterized in that the dictionary is registered in a user registration word dictionary referred to during speech recognition so that recognition processing can be performed using a model. 16. A speech uttered by a user for word registration is converted into a sequence of at least one partial word, and for each of the converted partial word sequences, information representing the partial word sequence is used by the user. And accepts the designation from the user as to whether or not the registration is possible, and confirms whether or not the registration of the corresponding partial word series is possible in accordance with the accepted specification contents. Information corresponding to the partial word sequence for which registration has been confirmed is obtained, and a word voice model in which partial word voice models are joined from information corresponding to the partial word sequence is obtained, and recognition processing can be performed using the word voice model. A method for creating a word dictionary for subword-type unspecified speaker speech recognition, characterized in that the dictionary is registered in a user registration word dictionary referred to during speech recognition. 17. A speech uttered by a user for word registration is converted into a sequence of at least one partial word, and for each of the converted partial word sequences, information representing the partial word sequence is used by the user. And accepts the user's editing operation for the information, performs the editing process on the information, and reflects the result of the editing process on the corresponding partial word sequence. Upon receiving the specification from the user, confirming whether or not registration of the corresponding partial word sequence is possible according to the received specification content. When confirming the registration possibility, the partial word sequence in which the registration is confirmed is confirmed. A word speech model in which partial word speech models are joined from information corresponding to the partial word sequence is obtained from information corresponding to the partial word sequence, and recognition processing is performed using the word speech model. A method for creating a word dictionary for subword-type unspecified speaker voice recognition, characterized in that the word dictionary is registered in a user-registered word dictionary that is referred to during voice recognition so as to perform voice recognition. 18. A speech uttered by a user in a word registration mode is converted into a sequence of at least one partial word, and information on the converted partial word sequence is registered in a user registration word dictionary. In the recognition mode, a word speech model in which partial word speech models are joined is acquired from information corresponding to each partial word sequence registered in the user registered word dictionary, and a speech uttered by the user is acquired. A speech recognition method characterized by performing recognition using a word speech model. 19. A speech uttered by a user in a word registration mode is converted into at least one partial word sequence, and it is determined whether or not the converted partial word sequence satisfies a predetermined registration condition. Determining, when it is determined that the registration condition is satisfied, the converted partial word sequence information is registered in the user registration word dictionary. In the voice recognition mode, the information is registered in the user registration word dictionary. A word speech model in which partial word speech models are joined is obtained from information corresponding to each partial word sequence, and a voice uttered by a user is recognized using the obtained word speech model. Voice recognition method. 20. A speech uttered by a user in a word registration mode is converted into a sequence of at least one partial word, and for all of the converted partial word sequences, information representing the partial word sequence is provided to the user. Presenting, accepting a designation from the user as to whether or not the registration is possible, confirming whether or not registration of the corresponding partial word sequence is possible in accordance with the accepted designation content, and confirming whether or not the registration is possible, In the voice recognition mode, information corresponding to the partial word series for which permission has been confirmed is registered in the user registered word dictionary, and a part is extracted from the information corresponding to each partial word series registered in the user registered word dictionary. Speech recognition characterized by acquiring a word speech model in which word speech models are connected, and recognizing a speech uttered by a user using the acquired word speech model. Method. 21. A speech uttered by a user in a word registration mode is converted into a sequence of at least one partial word, and for all of the converted partial word sequences, information representing the partial word sequence is provided to the user. Present and accept the user's editing operation for the information, perform the editing process on the information, reflect the result of the editing process on the corresponding partial word sequence, and use the Accepting the specification from the user, confirming whether or not the registration of the corresponding partial word series is possible according to the received specification contents. Corresponding information is registered in the user registration word dictionary, and in the voice recognition mode, information corresponding to each partial word sequence registered in the user registration word dictionary Get the word speech model Luo partial word speech model is stitched together, the voice the user has uttered, speech recognition method, characterized by recognition using word speech model the acquired. 22. Recognizing a voice uttered by a user in a word registration mode, generating at least one partial word sequence, and registering the generated partial word sequence information in a user registration word dictionary. In the voice recognition mode, a word voice model in which partial word voice models are joined from information corresponding to each partial word sequence registered in the user registered word dictionary is acquired, and the voice uttered by the user is obtained. Recognizing by using the obtained word-speech model, outputting a recognition result and its likelihood, and recognizing a voice uttered by the user to recognize at least one partial word sequence and the likelihood of the sequence. Generating the partial word system based on at least one of the likelihood of the generated partial word sequence in the voice recognition mode, the recognition result, and the likelihood of the recognition result. Speech recognition method characterized in that the determining whether the registration, additional registration information of the partial word sequence to the user registered word dictionary in accordance with the determination result. 23. Recognizing a voice uttered by a user in a word registration mode, generating at least one partial word sequence, and registering the generated partial word sequence information in a user registration word dictionary. In the voice recognition mode, a word voice model in which partial word voice models are joined from information corresponding to each partial word sequence registered in the user registered word dictionary is acquired, and a voice uttered by the user is obtained. Recognizing using the obtained word voice model and outputting the recognition result,
Generating a sequence of at least one partial word by recognizing a voice uttered by the user, receiving a registration instruction from the user when outputting the recognition result, and receiving the registration instruction. A speech recognition method, wherein information corresponding to the generated and instructed partial word sequence is additionally registered in the user registered word dictionary. 24. A process of converting a voice uttered by a user for word registration into at least one partial word sequence, and converting information corresponding to the converted partial word sequence to the partial word sequence. Processing for acquiring a word speech model in which partial word speech models are joined from information to be performed and registering it in a user registration word dictionary referred to during speech recognition so that recognition processing can be performed using the word speech model. A computer-readable recording medium on which a program to be executed by a computer is recorded. 25. A process of converting a voice uttered by a user in a word registration mode into a sequence of at least one partial word; a process of registering information of the converted partial word sequence in a user registered word dictionary; In the voice recognition mode, a word voice model in which partial word voice models are joined from information corresponding to each partial word sequence registered in the user registered word dictionary is acquired, and a voice uttered by the user is obtained. A computer-readable recording medium in which a program for causing a computer to execute a process of recognizing using the acquired word voice model is recorded.