JP2001092494A - Device and method for recognizing speech, and speech recognition program recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To facilitate accurately giving a proper reading to a syllable string agreeing with an unknown word when the word is extracted during speech recognition processing. SOLUTION: This speech recognition device is provided with an unknown word syllable presuming device 8; which generates various syllable string candidates corresponding to the unknown word by referring to a sub-word dictionary 11 in which various readings to sub-words composing a word as syllable strings and combining the syllable strings to the sub-words composing the unknown word; which detects a syllable string candidate most approximate to the syllable string for recognition corresponding to the unknown word by referring to a table of differences 12 for evaluating a degree of approximation between two syllable strings; and thus, which presumes the maximum likelihood syllable string candidate as the syllable string agreeing with the unknown word.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to speech recognition, and more particularly to a speech recognition apparatus and method for extracting an unknown word from an utterance including an unknown word and estimating a syllable string of a reading that matches the unknown word. And a voice recognition program recording medium.

[0002]

2. Description of the Related Art Speech input is useful as a means for inputting Japanese documents, but it is desired to improve recognition accuracy for practical use. In particular, a method using a word dictionary has been considered in order to increase the recognition accuracy of input speech. However, in the method using the word dictionary, there is a limit to the number of words that can be registered in the word dictionary, and it is extremely difficult to correctly recognize a newly appearing word (unknown word). Therefore, as disclosed in, for example, JP-A-2-163874, when a user inputs a certified character string, a character string of an unknown word candidate is extracted using information such as a character type, and the unknown word by the user is extracted. Or an unknown word is identified by accessing a large-scale dictionary, and the unknown word is newly registered in the word dictionary.

FIG. 23 is a block diagram showing the configuration of a conventional general speech recognition apparatus having an unknown word extracting function.
In the figure, 101 is a microphone, 102 is a syllable string calculating device, 103 is a word string calculating device, 104 is an output device, 10
5 is a correction device, 106 is a character type division device, 107 is an unknown word extraction device, 108 is a RAM, and 109 is a word dictionary.

Next, the operation will be described. FIG. 24 is a flowchart showing a process of extracting an unknown word syllable using the above speech recognition device. When the user enters the microphone 101
, The process is started (step ST101). When a voice is input through the microphone 101 (step ST102), the input voice is converted into an electric signal inside the microphone 101 (step ST103). The syllable string calculating device 102 A / D converts and quantizes the electric signal, performs a spectrum analysis of the voice pattern, connects syllable unit recognition results to generate syllable string candidates, and stores the syllable string candidates in the RAM 108. It is stored (step ST104). The word string calculation device 103 calculates word string candidates corresponding to all syllable string candidates (step ST105). next,
The output device 104 selects the maximum likelihood syllable string candidate and word string candidate and outputs them (step ST106).
The user looks at the display output from the output device 104 and, if there is an error in the recognition result, corrects the erroneously recognized portion using the correction device 105 (step ST).
107). The correction device 105 receives the correction input from the user and outputs the correct character string. Then, the character type dividing device 106 inputs the correct character string output from the correction device 105 and converts the correct character string into character types (Hiragana, Katakana,
(Kanji, English characters, etc.), and outputs a character string divided into character types (step ST108). Unknown word extraction device 10
7 inputs a character type divided character string, searches the word dictionary 109 using a partial character string of the character type divided character string as a key, and when the character string as a key is not registered in the word dictionary, The resulting character string is output as an unknown word (step ST109). As described above, the unknown words included in the utterance are extracted, and the process is terminated (step ST11).
0).

[0005]

Since the conventional speech recognition apparatus is configured as described above, an unknown word can be extracted from the correct character string input by the user. Information on the syllable sequence of the reading that matches the unknown word is required. When an unknown word is written in katakana or hiragana, a syllable string can be given in some cases, but it is difficult to assign a syllable string with high accuracy to kanji or English character strings. As for the syllable string assignment for kanji, there has been proposed a method of connecting syllable strings corresponding to each single kanji and assigning an entire syllable string for an unknown word. There is a problem that it is difficult to select a correct syllable string with high accuracy because the syllable string candidates correspond.

[0006] Furthermore, for kana characters, for example, the rule of vowel lengthening is applied to a syllable string that matches the word "candle", and the matching syllable string is #roos.
oku #. However, if the above rule is similarly applied to a syllable string that matches the word “white rabbit”, the matching syllable string will be # siloosagi #, and a syllable string different from the correct syllable string # sirousagi # will be added. As described above, there is a problem that it is difficult to provide a syllable string with high accuracy by simply applying a syllable rule based on a simple notation even for a kana character.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and a speech recognition apparatus and a speech recognition apparatus capable of accurately giving an accurate syllable string matching an unknown word extracted in a speech recognition process. It is an object to obtain a recognition method and a recording medium for a speech recognition program.

[0008]

According to the present invention, there is provided a voice recognition apparatus comprising: a voice input unit configured to input a voice to generate an information-processable voice pattern represented by an electrical signal or the like; Syllable string calculating means for performing syllable unit recognition to calculate a syllable string candidate corresponding to the voice,
Word string calculating means for calculating a word string candidate corresponding to the syllable string candidate, and output means for outputting at least the maximum likelihood recognized word string calculated by the syllable string calculating means and the word string calculating means as a speech recognition result Correction means for a user to input a correct character string for correction when there is an error in a recognized word string displayed by the output means, and a morpheme for performing morphological analysis on the input correct character string Analyzing means, an unknown word range extracting means for comparing the correct character string and the morphological analysis result to identify an unknown word and a recognized syllable string corresponding to the unknown word, and reading various readings for subwords constituting the word into syllables. Generating various syllable string candidates corresponding to unknown words by combining syllable strings for subwords constituting unknown words with reference to subword dictionaries registered as strings A syllable string candidate that is most similar to the recognized syllable string corresponding to the unknown word is detected by referring to a difference table for evaluating the degree of approximation between the two syllable strings, and this maximum likelihood syllable string candidate is converted to an unknown word. An unknown word syllable estimating means for estimating a matching syllable string is provided.

A syllable string calculating means for performing syllable unit recognition based on a voice pattern to calculate a plurality of syllable string candidates having a higher likelihood corresponding to a voice, and Word string calculating means for calculating a word string candidate corresponding to each of the syllable string candidates; and a plurality of syllable string candidates and word string candidates calculated by the syllable string calculating means and the word string calculating means. Output means for detecting a combination having the largest linguistic likelihood from among the combinations of, and outputting at least a recognized word string as a syllable string candidate and a word string candidate for the combination as a recognized syllable string and a recognized word string, respectively. It is like that.

[0010] A speech recognition apparatus according to the present invention is a word registration for registering an unknown word recognized by an unknown word range extracting means and a syllable string matching the unknown word estimated by an unknown word syllable estimating means in a word dictionary. Means.

[0011] The speech recognition apparatus according to the present invention provides an unknown word recognized by the unknown word range extracting means and a syllable string matching the unknown word estimated by the unknown word syllable estimating means as n-gram in the word dictionary. N-gram to register
It is provided with registration means.

The speech recognition apparatus according to the present invention provides a user with a notation representing an unknown word recognized by an unknown word range extracting unit and a syllable string matching the unknown word estimated by the unknown word syllable estimating unit to a user. A second output means for displaying, and a second input means for inputting a notation that is correct when the notation representing the unknown word and the syllable string matching the unknown word displayed on the second output means is incorrect. 2 correction means.

The speech recognition apparatus according to the present invention registers a syllable string matching the unknown word estimated by the unknown word syllable estimating means in the word dictionary for the unknown word recognized by the unknown word range extracting means. In addition, a different notation registering means for registering the matching syllable string in the word dictionary with respect to the different notation of the unknown word is provided.

The speech recognition apparatus according to the present invention determines whether or not a syllable string matching the unknown word has been estimated for the unknown word recognized by the unknown word range extracting means. The estimated syllable string is registered in the word dictionary as a syllable string that matches the unknown word, and if it cannot be estimated, the recognized syllable string corresponding to the unknown word recognized by the unknown word range extracting means is converted to the unknown word. And a syllable string registering means for registering the syllable string in the word dictionary as a syllable string matching with.

The speech recognition apparatus according to the present invention registers a syllable string matching the unknown word estimated by the unknown word syllable estimating means in the word dictionary for the unknown word recognized by the unknown word range extracting means. And a different reading registration unit for registering, in the word dictionary, a syllable string of different reading corresponding to the unknown word.

A voice recognition method according to the present invention includes a voice input step of inputting a voice to generate an information-processable voice pattern represented by an electrical signal or the like, and performing syllable-based recognition based on the voice pattern. A syllable string calculating step of calculating a syllable string candidate corresponding to the voice, a word string calculating step of calculating a word string candidate corresponding to the syllable string candidate, and the syllable string calculating step as a speech recognition result. An output step of outputting at least the maximum likelihood recognized word string calculated in the word string calculation step, and a user inputting a correct character string for correction when the recognized word string displayed in the output step has an error A correcting step, a morphological analysis step of performing a morphological analysis on the input correct character string, and a morphological analysis result with the correct character string. An unknown word range extraction step of comparing an unknown word and a recognized syllable string corresponding to the unknown word, and referring to a subword dictionary in which various readings for the subwords constituting the word are registered as a syllable string. By generating syllable string candidates corresponding to the unknown word by combining syllable strings for the subwords constituting, the recognition corresponding to the unknown word is referred to by referring to a difference table for evaluating the degree of approximation between the two syllable strings. A syllable string candidate closest to the syllable string is detected, and an unknown word syllable estimation step of estimating the most likely syllable string candidate as a syllable string matching the unknown word is provided.

In the speech recognition method according to the present invention, in the syllable string calculation step, syllable unit recognition is performed on the basis of the speech pattern to calculate a plurality of syllable string candidates with higher likelihood corresponding to the speech, and the word string is calculated. In the calculating step, a word string candidate corresponding to each of the plurality of syllable string candidates is calculated, and in the outputting step, the syllable string calculating step and the plurality of syllable strings calculated in the word string calculating step are combined with the word string. A combination having the largest linguistic likelihood is detected from the combinations, and the syllable string candidate and the word string candidate related to the combination are output as at least a recognized syllable string and a recognized word string, respectively. .

In the speech recognition method according to the present invention, word registration for registering, in a word dictionary, an unknown word recognized in an unknown word range extraction step and a syllable string matching the unknown word estimated in the unknown word syllable estimation step. It has a step.

In the speech recognition method according to the present invention, in the word registration step, the unknown word recognized in the unknown word range extraction step and the syllable string matching the unknown word estimated in the unknown word syllable estimation step are n- gra
It is registered in the word dictionary as m.

In the speech recognition method according to the present invention, the notation representing the unknown word recognized in the unknown word range extraction step and the syllable string matching the unknown word estimated in the unknown word syllable estimation step is provided to the user. 2nd to display
And a second correction step in which the user inputs a correct notation when the notation representing the unknown word and the syllable string matching the unknown word displayed in the second output step is incorrect. It has the following.

In the speech recognition method according to the present invention, a syllable string matching the unknown word estimated in the unknown word syllable estimation step is registered in the word dictionary for the unknown word identified in the unknown word range extraction step. , A different notation registration step of registering the matching syllable string in the word dictionary even for the different notation of the unknown word.

The speech recognition method according to the present invention determines whether or not a syllable string matching the unknown word has been estimated for the unknown word identified in the unknown word range extracting step. The estimated syllable string is registered in the word dictionary as a syllable string that matches the unknown word, and if it cannot be estimated, the recognized syllable string corresponding to the unknown word identified in the unknown word range extraction step is replaced with the unknown word. Is registered in the word dictionary as a syllable string that matches

In the speech recognition method according to the present invention, a syllable string matching the unknown word estimated in the unknown word syllable estimation step is registered in the word dictionary for the unknown word identified in the unknown word range extraction step. , A misreading registration step of registering, in the word dictionary, a syllable string of misreading that matches the unknown word.

A syllable string calculation function for performing syllable unit recognition based on an input voice pattern to calculate a syllable string candidate corresponding to a voice, and A word string calculation function of calculating a word string candidate corresponding to the above, an output function of outputting at least the maximum likelihood recognized word string calculated using the syllable string calculation function and the word string calculation function, and the output function A correction function that allows a user to input a correct character string for correction when there is an error in a recognized word string displayed using
A morphological analysis function for performing morphological analysis on the input correct character string, and an unknown word for comparing the correct character string and the morphological analysis result to recognize an unknown word and a recognized syllable string corresponding to the unknown word Various syllable string candidates corresponding to unknown words by combining a range extraction function and syllable strings for subwords forming unknown words by referring to a subword dictionary in which various readings for subwords forming words are registered as syllable strings Is generated, and a syllable string candidate closest to the recognized syllable string corresponding to the unknown word is detected with reference to a difference table for evaluating the degree of approximation between the two syllable strings. A syllable sequence that matches an unknown word and an unknown word syllable estimation function that is recorded on a computer-readable recording medium with a speech recognition program for causing the computer to realize the function. That.

A syllable string calculating function for performing syllable unit recognition based on a voice pattern and calculating a plurality of syllable string candidates with higher likelihood corresponding to voice, according to the present invention. A word string calculation function for calculating a word string candidate corresponding to each of the plurality of syllable string candidates; and a plurality of syllable string candidates calculated using the syllable string calculation function and the word string calculation function. An output for detecting a combination having the largest linguistic likelihood from combinations with word string candidates and outputting at least a recognized word string as a syllable string and a recognized word string, respectively, as a syllable string candidate and a word string candidate relating to the combination; The functions are recorded on a computer-readable recording medium with a speech recognition program supplemented with a program for causing a computer to realize the functions.

[0026] The speech recognition program recording medium according to the present invention provides an unknown word recognized by using an unknown word range extraction function,
And a computer-readable recording medium recording a speech recognition program supplemented with a program for causing a computer to implement a word registration function of registering a syllable string matching the unknown word estimated by the unknown word syllable estimation function in a word dictionary. It was done.

[0027] The speech recognition program recording medium according to the present invention provides an unknown word recognized using an unknown word range extraction function,
And a speech recognition program that supplements a program for causing a computer to implement an n-gram registration function of registering a syllable string matching the unknown word estimated using the unknown word syllable estimation function as an n-gram in a word dictionary. It is recorded on a computer-readable recording medium.

[0028] The speech recognition program recording medium according to the present invention provides an unknown word recognized using an unknown word range extraction function.
And a second output function for displaying to the user a notation representing a syllable string that matches the unknown word estimated using the unknown word syllable estimation function, and the second output function displayed using the second output function. When a notation representing an unknown word and a syllable string that matches the unknown word has an error, a second correction function enabling a user to input a correct notation is added to a program for causing a computer to perform the correction. The voice recognition program is recorded on a computer-readable recording medium.

The speech recognition program recording medium according to the present invention provides a syllable string that matches an unknown word estimated using an unknown word syllable estimation function for an unknown word that has been identified using an unknown word range extraction function. A computer-readable speech recognition program that is supplemented with a program for realizing a computer to have a variant transcription registration function of registering the matching syllable string in the word dictionary with respect to a variant representation of an unknown word while registering the phrase in an unknown word. Recorded on a simple recording medium.

The speech recognition program recording medium according to the present invention is capable of determining whether or not a syllable string matching an unknown word has been estimated for an unknown word recognized using the unknown word range extraction function. If not, the estimated syllable string is registered in the word dictionary as a syllable string that matches the unknown word, and if it cannot be estimated, the recognition corresponding to the unknown word recognized using the unknown word range extraction function A computer-readable recording medium records a speech recognition program supplemented with a program for causing a computer to implement a syllable string registration function of registering a syllable string in a word dictionary as a syllable string matching the unknown word.

[0031] The speech recognition program recording medium according to the present invention provides a syllable string that matches the unknown word estimated using the unknown word syllable estimation function with respect to the unknown word recognized using the unknown word range extraction function. A computer-readable speech recognition program that supplements a program for realizing a variant reading registration function for registering in a word dictionary and registering, in the word dictionary, a variant syllable string that matches the unknown word with respect to the unknown word. It is recorded on a readable recording medium.

[0032]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a configuration of a speech recognition device according to Embodiment 1 of the present invention. In the figure, reference numeral 1 denotes a microphone (speech input means) which inputs a speech uttered by a user and converts it into an electric signal to generate a speech pattern which can be processed, and 2 denotes a speech pattern obtained by the microphone 1 based on the speech pattern. A syllable string calculating device (syllable string calculating means) for performing syllable unit recognition and calculating the maximum likelihood syllable string candidate corresponding to the speech, 3 is a word for calculating the maximum likelihood word string candidate based on the syllable string candidate A column calculating device (word string calculating means) 4 is an output device (output means) for outputting the maximum likelihood syllable string candidate and word string candidate as a recognized syllable string and a recognized word string as a speech recognition result. When there is an error in the recognition result displayed and output, a correction device (correction means) for receiving a correction input from the user for the erroneously recognized portion and outputting the correct character string; Positive after receiving input Morphological analysis apparatus for performing the morphological analysis on a character string (morphological analysis means), 7 morphological analysis apparatus 6
An unknown word range extracting device (an unknown word range extraction device) that recognizes an unknown word from the correct character string separated by the above and refers to a recognized syllable sequence output to the output device 4 to recognize a recognized syllable sequence corresponding to the unknown word Means) and 8 are unknown word syllable estimating devices (unknown word syllable estimating means) for estimating an accurate syllable string matching the unknown word recognized by the unknown word range extracting device 7.
Reference numeral 9 denotes a RAM for storing syllable string candidates and the like;
Is a word dictionary in which words are registered, 11 is a sub-word dictionary in which various readings for sub-words such as single kanji constituting the word are registered as syllable strings, and 12 is two syllables (syllable strings).
6 is a difference table showing the log likelihood between them.

The arithmetic processing performed by the word string calculating device 3 will be described. The word string candidate is to detect W that maximizes the probability P (W | Y) of the word string given by the following equation. Generated by

[0034]

(Equation 1)

In the above equation, W is the uttered word
The column Y indicates the uttered syllable sequence. P on the right side of the above equation
(Y) is the expected value of Y, which is independent of W
In order to find W that maximizes P (W | Y)
If we find W that maximizes P (Y | W) · P (W)
It will be good. Here, P (Y | W) is given by the word string W.
Is the probability of appearance of the syllable string Y when obtained, and P (W) is
This is the appearance probability of the word string W. Then, at time t = 1, 2, 2,
, L is a syllable string corresponding to the word string W, Y = Y₁, Y_Two, ..., Y_L  P (Y | W) is calculated from the syllable probability as
Therefore, it can be calculated.

[0036]

(Equation 2)

The appearance probability P (W) of the word string is m
A word string W consisting of the words₁, W_Two, ..., w_m  Is determined independently of the syllable probability as follows (word n−
(gram information).

[0038]

(Equation 3)

With the above-described calculation, for each syllable string candidate whose word string corresponding to the syllable string exists in the word dictionary 10, W that maximizes the word string probability P (W | Y) is calculated. . In the above formula, it is assumed that the appearance probability of each word is stored in the word dictionary 10 in advance. Also, as for the combination calculation, for example, V shown in Seiichi Nakagawa: “Speech Recognition by Stochastic Model”
It can be implemented at high speed using a method such as the iterbi method or the stack decoding method. Further, the probability may be set as a logarithmic probability, and the calculation formula may be calculated as a sum.

In the word dictionary 10, the character notation for the word, the syllable string notation of the reading that matches the word, and the appearance probability (likelihood) P (W) of the last word are expressed and stored as one record. ing. Table (a) shown in FIG.
Shows a storage example inside the word dictionary 10 using the storage form of 1-gram, and Table (b) shows a storage example inside the word dictionary 10 using the storage form of 2-gram. In this embodiment, word chains are handled up to two chains, but the number of chains may be three or more.

Next, the functions of the morphological analyzer 6, unknown word range extracting device 7, and unknown word syllable estimating device 8 will be described. The morphological analysis device 6 receives the correct character string given by the user as input, performs morphological analysis of the correct character string, divides the correct character string into morphemes, and outputs the morpheme. Here, the morpheme is composed of three elements: morpheme character notation, morpheme syllable string notation, and morpheme part of speech. By the morphological analysis, for example, a morpheme such as “speech-onsee noun” is divided and output.

The unknown word range extraction device 7 receives the morphological analysis result and the recognized syllable sequence and the recognized word sequence displayed and output by the output device 4, compares the morpheme with the recognized word sequence, and differs in character notation. A morpheme including a character string is set as an unknown word candidate. Next, the word dictionary 10 is searched using the character string given from the character notation of the morpheme of the unknown word candidate as a key. If the character string is not registered in the word dictionary 10, the unknown word candidate is recognized as an unknown word. At the same time, a partial recognized syllable string corresponding to the unknown word in the recognized syllable string output to the output device 4 is recognized and output as an unknown word range. In this embodiment, the morpheme and the recognized word string are compared, and a morpheme including a character string having a different character notation is determined as an unknown word candidate. However, a subword or a series of subword strings may be determined as an unknown word candidate. As in the conventional method, words separated by character types such as kana and kanji may be obtained as unknown word candidates.

The unknown word syllable estimating device 8 estimates a correct syllable sequence matching the unknown word based on the unknown word range input from the unknown word range extracting device 7. At this time, the unknown word syllable estimation device 8 divides the character string given as the unknown word into an arbitrary partial character string using the subword dictionary 11, and calculates the syllable string given to the subword obtained by the division. In various combinations, all syllable string candidates corresponding to unknown words are generated. Then, for all the syllable string candidates obtained in this way, the similarity between the syllable string candidates recognized by the unknown word range extraction device 7 and the recognized syllable strings corresponding to the unknown words is calculated by using the difference table 12.
And output the syllable string with the highest degree of approximation as a syllable string that matches the reading of the unknown word.

Here, the sub-word dictionary 11 registers syllable strings of various readings for the sub-words constituting the word in order to add syllable strings matching unknown words not registered in the word dictionary 10. is there. For example, syllable strings for reading various single kanji characters constituting words, syllable strings for reading subwords when various pronunciation rules such as vowel lengthening rules or the like are applied or not are registered. FIG. 3 shows an example of data registered in the sub-word dictionary 11.

As shown in FIG. 4, the difference table 12 shows the log likelihood between two contrasted syllables (or syllable strings). Therefore, the sub-word dictionary 11
When the syllable string candidates generated by referring to the above are compared with the recognized syllable string corresponding to the unknown word, each syllable string is appropriately divided, and each corresponding partial syllable (syllable string)
Is determined, and the log likelihood between the corresponding syllables obtained from the difference table 12 is summed to calculate the degree of approximation between the syllable string candidate and the recognized syllable string.

Next, the operation of the speech recognition apparatus according to Embodiment 1 of the present invention will be described with reference to specific examples.
FIG. 5 is a flowchart showing a process of extracting unknown word syllables using the speech recognition device according to the first embodiment of the present invention. The process is started when the user speaks to the microphone 1 (step ST1). When a voice is input through the microphone 1 (step ST2), the input voice is converted into an electric signal inside the microphone 1 and taken in as analog data. (Step ST3). In the present embodiment, it is assumed that the user has uttered “Onsenishoshi”.

The syllable string calculating device 2 A / D converts and quantizes the analog data taken in by the microphone 1, and then performs a spectrum analysis to separate each syllable candidate recognized in syllable units. The connection is made and one maximum likelihood syllable string candidate is output (step ST4). For various methods for selecting syllable string candidates, see, for example, Seichi Nakagawa:
The details are disclosed in "Speech Recognition by Probability Model". In this embodiment, the following maximum likelihood syllable string candidates and their likelihoods are output. # ONseeniNhisiriri # 0.3 where # is a symbol representing the beginning and end of a sentence. In addition,
For the acoustic likelihood, a logarithmic probability or the like may be used instead of the probability.

The word string calculator 3 calculates word string candidates based on the maximum likelihood syllable string candidates output from the syllable string calculator 2 (step ST5). Abe et al .:
A two-stage search method taking into account the difference tendency between the first-stage optimal solution and the correct solution is used, a method described in Tone Composition Theory, 1-R-15, 1998. At this time, on the assumption that the maximum likelihood syllable string candidate calculated by the syllable string calculating device 2 is the only syllable string candidate, the most likely word string candidate corresponding to the syllable string candidate is detected, and The maximum likelihood word string candidate and its likelihood are output. "Speech recognition processing" 0.4

Then, the maximum likelihood syllable string candidates and word string candidates are stored in the RAM 9 as recognized syllable strings and recognized word strings (step ST6). # NULL Sentence / voice onsee noun / cognition
niNhi noun / processing sjori saga noun / #
NUL sentence end probability 0.4

Next, the output device 4 reads out the maximum likelihood recognized word string stored in the RAM 9, and connects the notation elements other than # to output (step ST7). In this embodiment, the following character strings are output. "Speech recognition processing"

When there is an error in the recognized word string displayed and output by the output device 4, the user inputs a correct character string using the correction device 5 (step ST8). Here, if there is no error in the character string and there is no need for correction by the user, the process ends. In this embodiment, a correct character string shown below is input. "Speech Recognition Processing" FIG. 6 shows a character string before correction and a character string after correction.

Next, the morphological analyzer 6 performs a morphological analysis on the correct character string that becomes the correct answer (step ST9). In this embodiment, the morphological analysis result is as follows. # NULL Sentence / voice onsee noun / recognition
? Noun / processing sjori sa-variant noun / # NULL sentence end

Here, the morphological analysis algorithm will be described in detail. FIG. 7 is a flowchart showing an algorithm of morphological analysis. First, the correct character string “speech recognition process”, which is the correct answer, is input using the correction device 5 (step ST121). Next, the morphological analyzer 6
The input correct character string is stored in the RAM 9 (step ST122). At this time, the correct character string is decomposed into a combination of subwords using the subword dictionary 11. Here, the subwords "sound", "voice", "recognition",
Given "insight", "action" and "physics", construct a virtual word for any combination of these subwords,
It is assumed that a probability is added to the word length extracted in advance from a large amount of corpus and stored in the RAM 9. Here, it is assumed that “recognition” is an unknown word that is not registered in the word dictionary 10.

The morphological analyzer 6 stores the correct character string in the RAM
9 and an initialization process is performed (step ST123). As initialization processing, the null word "#
"# Sentence head" and its probability value "1" are stored in the RAM 9 as initial values of the preceding word string.

Next, the morphological analysis device 6 searches the preceding word string by referring to the word dictionary 10 and the subword dictionary 11 until the preceding word string matches the correct character string by searching for a character string having a prefix match. Constitute. In this embodiment, first, “# # sentence head” is extracted as a preceding word string (step ST124).

When the preceding word string is set, it is checked whether or not there is a backward word that matches forward in the partial character string after the preceding word string in the correct character string (step ST125). Regarding the search for the backward word, the word dictionary 10 and the subword dictionary 11 are referred to and treated as the backward word as long as the subword and the subword string are also registered. Here, if there is no backward word that matches forward, the process returns to step ST124 to initialize the preceding word string. Also,
If there is a backward word that matches the head, the likelihood of the matched backward word is calculated and stored in the RAM 9, and the backward word is connected to the preceding word string and stored in the RAM 9 as a new preceding word string. (Step ST12
6). In the present embodiment, in order to search for a word after the preceding word string “# # sentence head”, whether a matching character string exists in the word dictionary 10 in “speech recognition processing #” following “#” Search for no. In this case, since "voice" is registered in the word dictionary 10, "voice onsee"
Nouns ”are extracted as backward words. Then, the preceding word string “# # sentence head” is changed to “# # sentence head voice onsee.
Noun ". Here, it is assumed that the probability of 2-gram is used for the calculation of the language likelihood, and "####
The linguistic likelihood for the sentence head voice onsee noun is
The probability of the preceding word string “## sentence head” (that is, “1”) and “## sentence head sound o described in the word dictionary 10
nsee noun "is given as the product of 2-gram probability.

Next, it is checked whether or not the preceding word string matches the correct character string (step ST127). If the preceding word string does not match the correct character string, the process proceeds to step ST1 to detect further backward words.
Return to 25. Also, if the preceding word string matches the correct character string, if there is already a preceding word string that matches the correct character string by another combination of subwords, the preceding word string with the maximum linguistic likelihood is determined. Upon detection, the preceding word string and its likelihood are stored in the RAM 9 (step ST128).

Next, it is checked whether or not a search has been performed for all combinations of preceding word strings (step S).
T129). If the search has not been performed for all combinations, the process returns to step ST124 to detect a preceding word string composed of other combinations and calculate its linguistic likelihood. If the search has been completed for all combinations, the combination of preceding word strings having the maximum linguistic likelihood stored in the RAM 9 is read out, and the combination is output as a morphological analysis solution ( Step ST130). In the present embodiment, for the correct character string "#voice recognition processing #", "# # sentence head voice onsee noun recognition?"
Noun processing sjori sa noun # # sentence end "
Is output as the analysis result of the morphological analysis. On this occasion,
"Recognition" which is an unknown word not registered in the word dictionary 10
About syllable string notation unknown in morphological analysis
Is output as Then, the process returns to step ST9 with the morpheme sequence of the solution as a return value (step ST13).
1).

When the process returns to step ST 9, the unknown word range extraction device 7 compares the recognized word sequence “speech recognition process” output to the output device 4 with the result of the morphological analysis, (That is, a morpheme including a corrected part determined by "knowledge->knowledge"), and an unknown word "recognition" from the recognition syllable string # oNseeniNhisjori # output to the output device 4. "Is partially recognized syllable string #niN
hi # is detected, and the unknown word "recognition" and the recognized syllable string # niNhi # corresponding to the unknown word are recognized and output as the unknown word range (step ST10).

Next, the unknown word syllable estimating device 8 analyzes the input unknown word range using the difference table 12, and estimates an accurate syllable string corresponding to the unknown word (step ST11). . At this time, in order to obtain a syllable string of accurate reading that matches the unknown word “recognition”, first, the subword dictionary 11
Search for. In the sub-word dictionary 11, "authorization" and "knowledge" are registered as sub-words.
, There is a reading of # niN # and # mitome #, and for "knowledge", it is found that there is # siki #. These combinations # niNsiki # and #mi
Calculate the degree of approximation between tomeshiki # and # niNhi #. In calculating the degree of approximation, as shown in FIG. 8, the two syllable strings to be compared are appropriately divided into syllable units so that the sum of the log likelihoods of the corresponding syllables is maximized, and the syllable level is adjusted. , And the log likelihood of each associated syllable pair is derived from the difference table shown in FIG. 4, and an approximation is given as a sum of these log likelihoods. In the difference table 12, usually, the optimal syllable string is given as a syllable string recognized by the speech recognition device,
The standard syllable sequence is given as a correct syllable sequence. Then, # niNsiki # is estimated as a syllable string that matches the unknown word, and the process ends (step ST1).
2).

As described above, according to the first embodiment, the speech recognition apparatus includes the morphological analyzer 6, the unknown word range extracting device 7, the unknown word syllable estimating device 8, the subword dictionary 11, and the difference table 12. Thus, the unknown word range extraction device 7 is used to compare the morphological analysis result of the morphological analysis device 6 with the recognized word sequence and the recognized syllable sequence output to the output device 4 to correspond to the unknown word and the unknown word. A recognized syllable string to be recognized is identified, and by referring to the subword dictionary 11, the degree of approximation between various syllable string candidates generated by combining subwords constituting an unknown word and the recognized syllable string corresponding to the unknown word is calculated. Since the syllable string candidate with the highest degree of approximation is estimated as a syllable string that matches the unknown word, an accurate syllable string can be accurately assigned to the unknown word extracted in the speech recognition process. The effect say.

The syllable string calculating device 2, word string calculating device 3, output device 4, correcting device 5, morphological analyzer 6, unknown word range extracting device 7, and unknown word syllable estimating device which constitute the speech recognition device. The function provided by 8 is CP
It can be realized by a program operating on a computer including a U, a memory, an input / output device, and the like.
Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 2 FIG. 9 is a block diagram showing a configuration of a speech recognition device according to Embodiment 2 of the present invention. 9, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will not be repeated. In the second embodiment, as compared with the first embodiment, the top N syllable string candidates having a higher likelihood are output, and a word string candidate corresponding to each syllable string candidate is calculated. A syllable string candidate and a word string candidate giving the largest linguistic likelihood in combination with a word string candidate are recognized syllable strings and a recognized word string, and a syllable string matching an unknown word is estimated based on these. . In FIG. 9, 21 is the microphone 1
N best syllable string calculating device (syllable string calculating means) for performing syllable unit recognition based on the voice pattern obtained by the above and outputting the top N syllable string candidates with high likelihood, 22 is the N best syllable string N that outputs the maximum likelihood word string candidate for each of the top N syllable string candidates output from calculation device 21
The best word string calculating device (word string calculating means) 23 is a syllable string that matches an unknown word based on the maximum likelihood recognized syllable string and the recognized word string from among combinations of N sets of syllable string candidates and word string candidates. Is the N best unknown word syllable estimation device.

Next, the operation will be described. FIG. 10 is a flowchart illustrating a process of estimating an unknown word syllable using the speech recognition device according to the second embodiment of the present invention. 10, the same reference numerals as those in FIG. 5 denote the same or corresponding processes, and a description thereof will not be repeated.

When the input speech is converted into an electrical signal in step ST3, the N best syllable string calculating device 21 outputs the top N syllable string candidates with high likelihood (step S3).
T21). Next, it is checked whether the calculation of word string candidates for all N syllable string candidates has been completed (step ST22). If the calculation of all word string candidates has been completed, the process proceeds to step ST7. Also,
If all word string candidates have not been calculated,
The maximum likelihood word string candidate is sequentially calculated for each syllable string candidate (step ST23). At this time, the product of the probability of occurrence of a syllable string candidate and the probability of occurrence of a word string candidate assuming the occurrence of the syllable string candidate is the language likelihood for the combination of the syllable string candidate and the corresponding word string candidate. Is calculated as For example, a syllable string candidate and its likelihood are #oNse
eniNhisjori #, 0.3, and if the maximum likelihood word string candidate and its likelihood are “speech recognition processing”, 0.4, assuming that the syllable string candidate is given, Syllable string candidate #oNseeniNhisjori
The linguistic likelihood for the combination of # and the word string candidate "speech recognition processing" is 0.12.

Next, it is checked whether or not the linguistic likelihood for the combination of the target syllable string candidate and word string candidate is maximum (step ST24). If it is not the maximum likelihood, the process returns to step ST22 to calculate a word string candidate corresponding to the next syllable string candidate. If it is the maximum likelihood, the syllable string candidate and the corresponding word string candidate are stored in the RAM 9. This allows
In step ST7, the stored syllable string candidates and word string candidates are read from the RAM 9, and the output device 4 displays and outputs at least the recognized word strings as the recognized syllable strings and the recognized word strings.

As described above, according to the second embodiment, the speech recognition apparatus is provided with the N best syllable string calculating device 21, the N best word string calculating device 22, and the N best unknown word syllable string estimating device 23. With this configuration, the N best word string calculating device 22 calculates corresponding word string candidates for the top N syllable string candidates calculated by the N best syllable string calculating device 21, and calculates the syllable string candidate and the word string candidate. A syllable string candidate and a word string candidate with the largest linguistic likelihood pertaining to the combination are regarded as a recognized syllable string and a recognized word string, and a syllable string matching the unknown word is estimated. Since the estimation is performed based on the total linguistic likelihood, syllable strings can be assigned to unknown words with higher accuracy.

The functions provided by the respective devices constituting the speech recognition device, such as the N-best syllable sequence calculating device 21, the N-best word sequence calculating device 22, and the N-best unknown word syllable sequence estimating device 23, are controlled by the CPU. , A memory, an input / output device, and the like. Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 3 FIG. 11 is a block diagram showing a configuration of a voice recognition device according to Embodiment 3 of the present invention. 11, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will not be repeated. Embodiment 3
Is different from Embodiment 1 in that an unknown word and a syllable string estimated to match the unknown word are registered in a word dictionary. In FIG. 11, reference numeral 31 denotes an unknown word and a syllable string estimated to match the unknown word.
Is a word registration device (word registration means).

Next, the operation will be described. FIG. 12 is a flowchart showing a process of estimating an unknown word syllable and registering the unknown word in a dictionary using the speech recognition device according to the third embodiment of the present invention. 12, the same reference numerals as those in FIG. 5 denote the same or corresponding processes, and a description thereof will not be repeated. In step ST11, for example, when the maximum likelihood syllable string # niNsiki # for the unknown word "recognition" is estimated, the word registration device 31 performs the unknown word character notation "recognition" and the unknown word syllabic string notation # niNsiki #. Is registered in the word dictionary 10 as a part of speech “noun”.

As described above, according to the third embodiment, the speech recognition device is configured to include the word registration device 31, so that the automatically recognized unknown word and the automatically estimated unknown word are determined. A syllable string that matches the word is a word dictionary 10
Therefore, there is an effect that the word dictionary 10 can be sequentially improved and the recognition accuracy can be improved.

The word registration device 31 and the like
The functions provided by each device constituting the speech recognition device can be realized by a program that operates on a computer including a CPU, a memory, an input / output device, and the like. Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 4 FIG. 13 is a block diagram showing a configuration of a voice recognition device according to Embodiment 4 of the present invention. 13, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will not be repeated. Embodiment 4
Is different from Embodiment 1 in that an unknown word and a syllable string estimated to match the unknown word are registered in a word dictionary as n-grams. In FIG. 13, 4
1 is an ng constructed by connecting an unknown word with a morpheme connected thereto
An n-gram registration device (n-gram registration means) for registering the ram in the word dictionary 10.

Next, the operation will be described. FIG. 14 is a flowchart showing a process of estimating an unknown word syllable by using the speech recognition apparatus according to the fourth embodiment of the present invention and registering the unknown word as an n-gram in a dictionary. FIG.
In FIG. 5, the same reference numerals as those in FIG. 5 indicate the same or corresponding processes, and a description thereof will not be repeated. In step ST11,
For example, the most likely syllable string # that matches the unknown word "recognition"
When the niNsiki # is estimated, the n-gram registration device 41 connects the unknown word to a morpheme connected to the unknown word, and
am, which is included in the n-gram, and the character notation “recognition” of the unknown word and the matching syllable string notation # niNsiki #
Is registered in the word dictionary 10 as a part of speech “noun”. At this time, a fixed value (for example, 0.1) is given as the language likelihood. For example, if it is registered in the word dictionary 10 in the form of 2-gram, the following records are registered in the word dictionary 10. Voice onsee noun recognition niNsiki noun 0.1 recognition niNsiki noun processing sjori sa noun 0.1

As described above, according to the fourth embodiment, since the voice recognition device is configured to have the n-gram registration device 41, the automatically recognized unknown words and the automatically estimated unknown words are obtained. The syllable string that matches the unknown word is n-
Since the words are registered in the word dictionary 10 in the form of a gram, the word dictionary 10 is successively enriched, and the target word can be accurately recognized based on words connected before and after the word, thereby improving the recognition accuracy. This has the effect that it can be performed.

The functions provided by the respective devices constituting the speech recognition device, such as the n-gram registration device 41, are realized by a program operating on a computer having a CPU, a memory, an input / output device, and the like. It is possible. Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 5 FIG. 15 is a block diagram showing a configuration of a speech recognition device according to Embodiment 5 of the present invention. 15, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will not be repeated. Embodiment 5
Presents to the user a syllable string that matches the unknown word recognized by the unknown word range extraction device and the unknown word estimated by the unknown word syllable estimation device, as compared with the first embodiment. The difference is that the matching syllable string can be modified. In FIG. 15, reference numeral 51 denotes a second output device (second output means) for displaying to the user the notation of the recognized unknown word and the estimated syllable string, and 52 denotes an unknown word or a syllable matching the unknown word. When there is an error in the sequence, a second correction device (second correction means) for inputting a correct character string or syllable sequence by the user for correction, the unknown word and a syllable sequence matching the unknown word are converted into words. A word registration device for registering in the dictionary 10.

Next, the operation will be described. FIG. 16 is a flowchart showing a process of extracting an unknown word using the voice recognition device according to the fifth embodiment of the present invention and registering it in a dictionary after correction by a user. In FIG.
5 denote the same or corresponding processes, and a description thereof will not be repeated. In step ST11, for example, a syllable string that matches the unknown word “recognition” is #mitom.
If esiki # is estimated, the second output device 51
Is a character notation “recognition” of an unknown word and a syllable string notation #mi
Tomesiki # is displayed and output, and the unknown word and the matching syllable string are presented to the user (step ST51).
Next, if there is an error in the output character representation or syllable string representation of the unknown word, the user corrects the character representation or syllable string representation to a correct character string or syllable string using the second correction device 52. I do. In this case, the syllable string #mitomes
iki # is modified to # niNsiki # (step S
T52). Then, the word registration device 53 registers the character notation “recognition” of the unknown word and the syllable string notation # niNsiki # in the word dictionary 10 as the part of speech “noun” (step ST5).
3).

As described above, according to the fifth embodiment, since the speech recognition device is configured to include the second output device 51 and the second correction device 52, the user can determine the unknown word and the matching syllable string. Since it is possible to confirm the error and correct the error if there is an error, an accurate voice data without any error can be surely registered in the dictionary, and the effect of improving the recognition accuracy can be obtained.

The functions provided by the respective devices constituting the speech recognition device, such as the second output device 51 and the second correction device 52, operate on a computer having a CPU, a memory, an input / output device and the like. It can be realized by a program that performs Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 6 FIG. FIG. 17 is a block diagram showing a configuration of a speech recognition device according to Embodiment 6 of the present invention. 17, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will not be repeated. Embodiment 6
Has the same reading for unknown words set as partial character strings in the correct character string input by the user, in addition to the regular character notation given by the user, as compared with the first embodiment. The difference is that syllable strings that match unknown words are registered in a word dictionary even for different character notations. In FIG. 17, reference numeral 61 denotes a syllable string that matches the same syllabic string that gives the same reading and meaning in addition to the regular letter notation for unknown words.
0 is a different notation registration device (different notation registration means).

Next, the operation will be described. FIG. 18 shows a process of extracting unknown words by using the speech recognition apparatus according to Embodiment 6 of the present invention and registering syllable strings that match not only regular expressions but also different expressions in the word dictionary 10. It is a flowchart shown. In FIG. 18, the same reference numerals as those in FIG. 5 denote the same or corresponding processes, and a description thereof will not be repeated.
In step ST11, assuming that a syllable string # kansu # that matches, for example, an unknown word “function” is estimated, the different notation registration device 61 converts the different notation character conversion rules (for example, Box)
Syllable string #kans that matches the above-mentioned "function"
The combination of uu # is also registered in the word dictionary 10.

As described above, according to the sixth embodiment, since the speech recognition apparatus is configured to have the different notation registration device 61, not only the regular notation input by the user for the unknown word but also the different notation is used. Since a syllable string that matches also is registered in the word dictionary 10 and learned, it is possible to improve the recognition accuracy.

The functions provided by each device constituting the speech recognition device, such as the different notation registration device 61, are realized by a program operating on a computer having a CPU, a memory, an input / output device, and the like. Is possible. Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 7 FIG. FIG. 19 is a block diagram showing a configuration of a speech recognition device according to Embodiment 7 of the present invention. In FIG. 19, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will be omitted. Embodiment 7
In comparison with the first embodiment, even if a syllable string matching an unknown word cannot be estimated, the dictionary recognizes a recognized syllable string corresponding to the unknown word recognized by the unknown word range extraction device as a syllable string matching the unknown word. It differs in that it is registered in. FIG.
In 71, it is determined whether a syllable string matching the unknown word has been estimated, and if it can be estimated, the estimated syllable string is registered in the word dictionary 10 as a syllable string corresponding to the unknown word. If it cannot be estimated, the syllable string registration device (syllable string registration means) registers the recognized syllable string corresponding to the unknown word recognized by the unknown word range extraction device 7 in the word dictionary 10 as a syllable string matching the unknown word. .

Next, the operation will be described. FIG. 20 is a flowchart showing a process of extracting an unknown word and registering a matching syllable string in a word dictionary using the speech recognition apparatus according to the seventh embodiment of the present invention. 20, the same reference numerals as those in FIG. 5 denote the same or corresponding processes, and a description thereof will not be repeated. After the syllable string that matches the unknown word is estimated in step ST11, it is checked whether the matching syllable string has been estimated (step ST71). If a matching syllable string can be estimated, the unknown word and the estimated syllable string are registered in the word dictionary 10 (step ST7).
3). When a matching syllable string cannot be estimated as in the case where the likelihood of a syllable string candidate for an unknown word is less than a predetermined threshold, the recognition syllable corresponding to the unknown word recognized by the unknown word range extraction device 7 The sequence is set as a syllable sequence that matches the unknown word (step ST72). Then, the unknown word and the set syllable string are registered in the word dictionary 10 (step ST73).

As described above, according to the seventh embodiment, the syllable string registration device 71 is provided.
Even when a partial character string not registered in the subword dictionary 11 exists in a character string extracted as an unknown word and a syllable string cannot be assigned to the recognized unknown word, a recognized syllable string corresponding to the unknown word Since the recognition syllable sequence is originally a recognition result for the utterance of the user, there is little possibility that a mismatch occurs in voice recognition, and the recognition accuracy can be improved.

The functions provided by the syllabic string registration device 71 and other devices constituting the speech recognition device are realized by a program operating on a computer having a CPU, a memory, an input / output device, and the like. Is possible. Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

Embodiment 8 FIG. FIG. 21 is a block diagram showing a configuration of a speech recognition device according to Embodiment 8 of the present invention. 21, the same reference numerals as those in FIG. 1 denote the same or corresponding parts, and a description thereof will not be repeated. Embodiment 8
Is compared with the first embodiment, the unknown word syllable estimation device 8
The difference is that, in addition to the syllable string estimated by, the syllable string of the misreading corresponding to the unknown word is registered in the word dictionary as matching the unknown word. In FIG. 21, in addition to the syllable string estimated by the unknown word syllable estimating device 8 with respect to the unknown word, the syllable string of misreading according to the rule for changing the syllable string matches the unknown word. Word dictionary 10
Is a different reading registration device (differential reading registering means).

Next, the operation will be described. FIG. 22 is a flowchart showing a process of extracting unknown words using the speech recognition apparatus according to Embodiment 8 of the present invention and registering, in addition to regular syllable strings, dyslexic syllable strings in the word dictionary 10. is there. 22, the same reference numerals as those in FIG. 5 denote the same or corresponding processes, and a description thereof will not be repeated. In step ST11, assuming that a syllable string # seNtakiki # that matches, for example, the unknown word “washing machine” is estimated, the misreading registration device 81 stores the syllable string conversion rules (eg, akuki->) held therein. Using aQki), syllable strings of different readings are also registered in the word dictionary 10. That is, "washing machine" and # seNtakiki #
Not only in combination with "washing machine" and #seNta
The combination with Qki # is also registered in the word dictionary 10.

As described above, according to the eighth embodiment, since the speech recognition device is configured to have the misreading registration device 81, the syllabic sequence of the misreading for an unknown word is automatically registered and learned. Therefore, there is an effect that recognition accuracy can be improved.

The functions provided by the devices constituting the speech recognition device, such as the different reading registration device 81, are realized by a program operating on a computer having a CPU, a memory, an input / output device, and the like. Is possible. Therefore, it is possible to record a program for realizing the above functions for performing the voice recognition processing on a computer-readable recording medium. Processing can be performed.

[0093]

As described above, according to the present invention, an unknown word and a recognized syllable string corresponding to the unknown word are recognized by comparing the correct character string input by the user with the result of the morphological analysis. Various syllable strings corresponding to unknown words are obtained by performing range extraction and combining syllable strings for subwords forming unknown words with reference to a subword dictionary in which various readings for subwords forming words are registered as syllable strings. A candidate is generated, a syllable string candidate closest to the recognized syllable string corresponding to the unknown word is detected with reference to a difference table for evaluating the degree of approximation between the two syllable strings, and this maximum likelihood syllable string candidate is determined. Since it is configured to perform unknown word syllable estimation to estimate a syllable string that matches the unknown word, it is possible to accurately assign an accurate syllable string to the unknown word extracted in the speech recognition process. An effect.

According to the present invention, syllable string calculation for performing syllable unit recognition based on a voice pattern to calculate a plurality of syllable string candidates with higher likelihood corresponding to voice, and A word string calculation for calculating a word string candidate corresponding thereto, and a combination having the largest linguistic likelihood is detected from combinations of a plurality of syllable string candidates and word string candidates, and a syllable string related to the combination is detected. Since the configuration is such that the candidate and the word string candidate are output as the recognized syllable string and the recognized word string, respectively, the estimation is performed based on the total linguistic likelihood including the combination of the syllable string and the word string. Therefore, there is an effect that it is possible to more accurately estimate a syllable string that matches an unknown word.

According to the present invention, since the word registration for registering the unknown word and the syllable string matching the unknown word in the word dictionary is performed, the automatically recognized unknown word and the automatically estimated unknown word are automatically registered. Since the syllable string that matches the unknown word is registered in the word dictionary, it is possible to enhance the word dictionary one by one to improve the recognition accuracy.

According to the present invention, n-gram registration is performed in which an unknown word and a syllable string matching the unknown word are registered in the word dictionary as n-grams. Since the syllable string that matches the unknown word automatically estimated is registered in the word dictionary in the form of n-gram, the word dictionary is sequentially enhanced, and the target word is determined based on the words connected before and after. Since it is possible to accurately recognize the object, it is possible to improve the recognition accuracy.

According to the present invention, a notation representing a syllable string matching an unknown word recognized by unknown word range extraction and an unknown word estimated by unknown word syllable estimation is displayed and output to a user. Since the user inputs the correct notation when there is an error in the notation relating to the word and the matching syllable string, the user can confirm the unknown word and the matching syllable string, and In the case where there is, this can be corrected, so that there is an effect that accurate voice data without any error can be reliably registered in the dictionary and the recognition accuracy can be improved.

According to the present invention, a syllable string matching the unknown word estimated by the unknown word syllable estimation is registered in the word dictionary for the unknown word identified by the unknown word range extraction, and the unknown word As for the notation, the syllable string that matches the syllabic string that matches not only the regular spelling entered by the user for the unknown word but also the syllable that matches the syllabic string is registered so that the matching syllable string is registered in the word dictionary. Since the columns are registered and learned in the word dictionary, the recognition accuracy can be improved.

According to the present invention, it is determined whether or not a syllable string matching the unknown word has been estimated for the unknown word identified by the unknown word range extraction. Register the string in the word dictionary as a syllable string that matches the unknown word, and if it cannot be estimated, register the recognized syllable string corresponding to the unknown word that has been certified by unknown word range extraction in the word dictionary as a syllable string that matches the unknown word Syllable string registration is performed, so a syllable string cannot be assigned to the recognized unknown word because there is a partial character string that is not registered in the subword dictionary in the character string extracted as an unknown word Even in this case, a recognition syllable sequence corresponding to the unknown word can be added, and since this recognition syllable sequence is originally a recognition result for the utterance of the user, there is little possibility that a mismatch occurs in speech recognition. An effect that it is possible to improve the recognition accuracy.

According to the present invention, a syllable string matching the unknown word estimated by the unknown word syllable estimation is registered in the word dictionary for the unknown word recognized by the unknown word range extraction. The system is configured to register the misreading syllable sequence that matches the unknown word in the word dictionary, so that the syllabic sequence of the misreading for the unknown word is automatically registered and learned, so that the recognition accuracy is improved. Is achieved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech recognition device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing the contents of a record stored in a word dictionary.

FIG. 3 is a diagram showing the contents of a record stored in a subword dictionary.

FIG. 4 is a diagram showing the contents of a record stored in a difference table.

FIG. 5 is a flowchart showing a speech recognition method according to the first embodiment of the present invention.

FIG. 6 is a diagram illustrating a recognized character string and a corrected character string.

FIG. 7 is a flowchart showing an algorithm of morphological analysis.

FIG. 8 is a diagram showing one process when calculating the degree of approximation between two syllable strings.

FIG. 9 is a block diagram showing a configuration of a voice recognition device according to a second embodiment of the present invention.

FIG. 10 is a flowchart showing a speech recognition method according to Embodiment 2 of the present invention.

FIG. 11 is a block diagram showing a configuration of a voice recognition device according to a third embodiment of the present invention.

FIG. 12 is a flowchart showing a voice recognition method according to Embodiment 3 of the present invention.

FIG. 13 is a block diagram showing a configuration of a voice recognition device according to a fourth embodiment of the present invention.

FIG. 14 is a flowchart showing a voice recognition method according to Embodiment 4 of the present invention.

FIG. 15 is a block diagram showing a configuration of a voice recognition device according to a fifth embodiment of the present invention.

FIG. 16 is a flowchart showing a voice recognition method according to Embodiment 5 of the present invention.

FIG. 17 is a block diagram showing a configuration of a speech recognition device according to Embodiment 6 of the present invention.

FIG. 18 is a flowchart showing a voice recognition method according to Embodiment 6 of the present invention.

FIG. 19 is a block diagram showing a configuration of a speech recognition device according to a seventh embodiment of the present invention.

FIG. 20 is a flowchart showing a voice recognition method according to Embodiment 7 of the present invention.

FIG. 21 is a block diagram showing a configuration of a voice recognition device according to an eighth embodiment of the present invention.

FIG. 22 is a flowchart showing a speech recognition method according to Embodiment 8 of the present invention.

FIG. 23 is a block diagram showing a configuration of a conventional speech recognition device having a general unknown word extraction function.

FIG. 24 is a diagram showing an operation process of extracting an unknown word syllable using a conventional speech recognition device.

[Explanation of symbols]

Reference Signs List 1 microphone (voice input means), 2 syllable string calculating device (syllable string calculating means), 3 word string calculating device (word string calculating means), 4 output device (output means), 5 correcting device (correcting means), 6 morpheme Analysis device (morphological analysis means), 7
Unknown word range extracting device (unknown word range extracting means), 8 unknown word syllable estimating device (unknown word syllable estimating means) 9 RAM, 1
0 word dictionary, 11 sub-word dictionary, 12 difference table,
21 N best syllable string calculating device (syllable string calculating means), 2
2 N best word string calculating device (word string calculating means), 23
N best unknown word syllable estimation device, 31 word registration device (word registration means), 41 n-gram registration device (n-
(gram registration means), 51 second output device (second output means), 52 second correction device (second correction means), 53
Word registration device, 61 Different notation registration device (different notation registration means), 71 Syllable string registration device (syllable string registration means), 81
Different reading registration device (different reading registration unit).

Claims (24)

[Claims] 1. A voice input means for inputting a voice to generate an information-processable voice pattern represented by an electric signal or the like, and performing syllable-based recognition based on the voice pattern to correspond to the voice A syllable string calculating means for calculating a syllable string candidate to be performed, a word string calculating means for calculating a word string candidate corresponding to the syllable string candidate, and a syllable string calculating means and a word string calculating means calculated as a speech recognition result. Output means for outputting at least the maximum likelihood recognized word string, and correction means for the user to input a correct character string for correction when there is an error in the recognized word string displayed by the output means,
A morphological analysis unit that performs morphological analysis on the input correct character string, and an unknown word that compares the correct character string with the result of the morphological analysis to identify an unknown word and a recognized syllable string corresponding to the unknown word. By combining a range extracting means and a syllable sequence for a subword constituting an unknown word with reference to a subword dictionary in which various readings for a subword constituting a word are registered as a syllable sequence, various syllable sequence candidates corresponding to the unknown word are obtained. A syllable string candidate that is most similar to the recognized syllable string corresponding to the unknown word is detected with reference to a difference table that evaluates the degree of approximation between the two syllable strings. A speech recognition apparatus comprising: an unknown word syllable estimating unit that estimates a syllable string that matches a word. 2. A syllable string calculating means for performing syllable unit recognition based on a voice pattern to calculate a plurality of syllable string candidates having a higher likelihood corresponding to the voice, and each of the plurality of syllable string candidates. A word string calculating means for calculating a word string candidate corresponding to the syllable string calculating means and the largest among the combinations of the plurality of syllable string candidates and the word string candidates calculated by the word string calculating means. Output means for detecting a combination having linguistic likelihood, and outputting at least a recognized word string as a syllable string candidate and a word string candidate relating to the combination as a recognized syllable string and a recognized word string, respectively. 2. The speech recognition device according to 1. 3. A word registering means for registering, in a word dictionary, an unknown word recognized by the unknown word range extracting means and a syllable string matching the unknown word estimated by the unknown word syllable estimating means. The voice recognition device according to claim 1. 4. An n-gram registration means for registering an unknown word recognized by the unknown word range extracting means and a syllable string matching the unknown word estimated by the unknown word syllable estimating means as an n-gram in a word dictionary. The voice recognition device according to claim 3, comprising: 5. A second output means for displaying, to a user, a notation representing a syllable string matching the unknown word recognized by the unknown word range extracting means and the unknown word estimated by the unknown word syllable estimating means. And a second correction unit for inputting a notation that allows the user to enter a correct answer when there is an error in the notation representing the unknown word and the syllable string matching the unknown word displayed on the second output unit. The speech recognition device according to claim 3 or 4, wherein: 6. A syllable sequence that matches the unknown word estimated by the unknown word syllable estimating means is registered in the word dictionary for the unknown word recognized by the unknown word range extracting means, and the unknown notation of the unknown word is registered. 2. The speech recognition apparatus according to claim 1, further comprising a different notation registration unit that registers the matching syllable string in a word dictionary. 7. It is determined whether or not a syllable string matching the unknown word can be estimated for the unknown word recognized by the unknown word range extracting means. If the syllable string can be estimated, the estimated syllable string is replaced with the estimated syllable string. The syllable string corresponding to the unknown word is registered in the word dictionary as a syllable string matching the unknown word, and if it cannot be estimated, the recognized syllable string corresponding to the unknown word recognized by the unknown word range extracting means is set as a syllable string matching the unknown word. The speech recognition apparatus according to claim 1, further comprising a syllable string registration unit that registers the syllable string in a dictionary. 8. A syllable string that matches the unknown word estimated by the unknown word syllable estimating means for the unknown word recognized by the unknown word range extracting means is registered in a word dictionary, and 2. The speech recognition apparatus according to claim 1, further comprising: a misreading registration unit that registers a misreading syllable string matching the unknown word in a word dictionary. 9. A voice inputting step of inputting a voice to generate an information-processable voice pattern represented by an electric signal or the like, and performing syllable-based recognition based on the voice pattern to correspond to the voice A syllable string calculating step of calculating a syllable string candidate to be performed, a word string calculating step of calculating a word string candidate corresponding to the syllable string candidate, and a syllable string calculating step and a word string calculating step calculated as a speech recognition result. An output step of outputting a recognized word string of at least the maximum likelihood, and a correction step of a user inputting a correct character string for correction when there is an error in the recognized word string displayed in the output step. A morphological analysis step of performing a morphological analysis on the correct character string, and comparing the correct character string with the morphological analysis result to determine whether the unknown word Combining an unknown word range extraction step of identifying a recognized syllable string corresponding to an unknown word with a syllable string for a subword forming an unknown word by referring to a subword dictionary in which various readings for the subwords forming the word are registered as a syllable string By generating various syllable string candidates corresponding to the unknown word, and referring to a difference table for evaluating the degree of approximation between the two syllable strings, a syllable string candidate that is the closest to the recognized syllable string corresponding to the unknown word And an unknown word syllable estimation step of estimating the maximum likelihood syllable string candidate as a syllable string matching the unknown word. 10. In the syllable string calculation step, syllable unit recognition is performed based on the voice pattern to calculate a plurality of syllable string candidates with higher likelihood corresponding to the voice, and in the word string calculation step, the plurality of syllable strings are calculated. A word string candidate corresponding to each of the string candidates is calculated, and in the output step, the largest language is selected from the syllable string calculating step and the combination of the plurality of syllable strings and the word strings calculated in the word string calculating step. 10. The speech recognition according to claim 9, wherein a combination having likelihood is detected, and at least a recognition word string is output as a syllable string candidate and a word string candidate relating to the combination, respectively, as a recognition syllable string and a recognition word string. Method. 11. A word registration step of registering, in a word dictionary, an unknown word identified in the unknown word range extraction step and a syllable string matching the unknown word estimated in the unknown word syllable estimation step. The speech recognition method according to claim 9. 12. In the word registration step, the unknown word recognized in the unknown word range extraction step and the syllable string matching the unknown word estimated in the unknown word syllable estimation step are registered in the word dictionary as n-grams. The speech recognition method according to claim 11, wherein: 13. A second output step of displaying, to a user, a notation representing a syllable string that matches the unknown word recognized in the unknown word range extraction step and the unknown word estimated in the unknown word syllable estimation step. And a second correction step in which the user inputs a correct notation when there is an error in the notation representing the unknown word and the syllable string matching the unknown word displayed in the second output step. The speech recognition method according to claim 11 or 12, wherein: 14. A syllable sequence that matches the unknown word estimated in the unknown word syllable estimation step is registered in the word dictionary for the unknown word identified in the unknown word range extraction step, and a difference between the unknown word and the unknown word is determined. 10. The speech recognition method according to claim 9, further comprising a different notation registration step of registering the matching syllable string in a word dictionary for the notation. 15. For the unknown word identified in the unknown word range extraction step, it is determined whether or not a syllable string matching the unknown word has been estimated. If the syllable string has been estimated, the estimated syllable string is determined. The syllable string corresponding to the unknown word is registered in the word dictionary as a syllable string matching the unknown word, and if it cannot be estimated, the recognized syllable string corresponding to the unknown word recognized in the unknown word range extracting step is set as a syllable string matching the unknown word. 10. A syllable string registration step for registering in a dictionary.
Voice recognition method described in. 16. A syllable string matching the unknown word estimated in the unknown word syllable estimation step is registered in the word dictionary for the unknown word recognized in the unknown word range extraction step, and the unknown word is A speech recognition method, comprising a misreading registration step of registering a misreading syllable string matching the unknown word in a word dictionary. 17. A syllable string calculating function for performing syllable unit recognition based on an input voice pattern to calculate a syllable string candidate corresponding to a voice, and calculating a word string candidate corresponding to the syllable string candidate. A word string calculation function, an output function of outputting at least the maximum likelihood recognized word string calculated using the syllable string calculation function and the word string calculation function, and a recognized word string displayed using the output function. A correction function that enables a user to input a correct character string for correction when there is an error; a morphological analysis function that performs morphological analysis on the input correct character string; and a correct character string and a morpheme An unknown word range extraction function for comparing an analysis result with an unknown word and a recognized syllable string corresponding to the unknown word, and a subword in which various readings for subwords constituting the word are registered as a syllable string. A syllable sequence corresponding to the unknown word is generated by combining syllable sequences for subwords constituting the unknown word with reference to the word dictionary, and a difference table for evaluating the degree of approximation between the two syllable sequences is referred to. An unknown word syllable estimation function for detecting a syllable string candidate closest to the recognized syllable string corresponding to the unknown word and estimating the maximum likelihood syllable string candidate as a syllable string matching the unknown word. A computer-readable recording medium on which a speech recognition program to be realized is recorded. 18. A syllable string calculating function for performing syllable unit recognition based on a voice pattern to calculate a plurality of syllable string candidates having a higher likelihood corresponding to a voice, and each of the plurality of syllable string candidates. A word string calculation function for calculating a corresponding word string candidate, and a combination of a plurality of syllable string candidates and word string candidates calculated using the syllable string calculation function and the word string calculation function. An output function of detecting a combination having the largest language likelihood and outputting at least a recognized word string as a syllable string candidate and a word string candidate for the combination as a recognized syllable string and a recognized word string,
18. A computer-readable recording medium recording a speech recognition program according to claim 17, wherein a program to be realized by a computer is additionally recorded. 19. A computer has a word registration function of registering, in a word dictionary, an unknown word recognized using an unknown word range extraction function and a syllable string matching the unknown word estimated by the unknown word syllable estimation function. 18. A computer-readable recording medium on which a speech recognition program according to claim 17, further comprising a program for causing the computer to execute the program. 20. Registering an unknown word recognized using an unknown word range extraction function and a syllable string matching the unknown word estimated using an unknown word syllable estimation function as an n-gram in a word dictionary. 20. The computer-readable recording medium according to claim 19, further comprising a program for causing a computer to realize a gram registration function. 21. A method for displaying, to a user, a notation representing a syllable string that matches an unknown word recognized using an unknown word range extraction function and the unknown word estimated using an unknown word syllable estimation function. And the user can input a correct notation when the notation representing the unknown word and the syllable string that matches the unknown word displayed using the second output function is incorrect. 21. A computer-readable recording medium storing a speech recognition program according to claim 19, wherein a program for realizing the second correction function to be executed by a computer is additionally recorded. 22. A syllable string matching the unknown word estimated using the unknown word syllable estimation function is registered in the word dictionary with respect to the unknown word recognized using the unknown word range extraction function. 18. The speech recognition according to claim 17, wherein a program for causing a computer to realize a function of registering a different syllabic string in the word dictionary for the different spelling of the word is additionally recorded. A computer-readable recording medium on which a program is recorded. 23. For unknown words recognized using the unknown word range extraction function, it is determined whether or not a syllable string that matches the unknown words has been estimated. If the syllable string can be estimated, the estimated syllables are determined. The sequence is registered in the word dictionary as a syllable sequence that matches the unknown word, and if it cannot be estimated, the recognized syllable sequence corresponding to the unknown word recognized using the unknown word range extraction function matches the unknown word. 18. The computer-readable recording medium according to claim 17, wherein a program for causing a computer to realize a syllable string registration function of registering a syllable string in a word dictionary is additionally recorded. 24. A syllable sequence that matches an unknown word estimated using an unknown word syllable estimation function for an unknown word identified using an unknown word range extraction function is registered in a word dictionary, and the unknown word is registered in the word dictionary. 18. The program according to claim 17, further comprising a program for causing a computer to realize a misreading registration function of registering, in a word dictionary, a syllable string of misreading that matches the unknown word for the word. A computer-readable recording medium on which a voice recognition program is recorded.