JPS5961897A - Recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Technical Field] The present invention relates to an improvement in a recognition device, and more specifically, the present invention relates to an improvement in a recognition device, and more specifically, for example, a recognition device that recognizes a section of information such as a section of speech such as a phrase, etc. This invention relates to the improvement of a recognition device that recognizes subdivided unit elements.

<Prior art> When recognizing one segment of speech, such as a phrase, in more subdivided units such as phonemes, kana, syllables, etc., conventionally, the input speech information of one segment to be recognized is generally processed through acoustic processing, for example. Then, a feature vector input cover pattern is obtained for each unit such as phoneme, syllable, etc., and this input cover pattern is matched with a pre-stored standard pattern, and the input information is used as a candidate unit sequence to select a pattern with high similarity. The output candidate unit string is compared with the contents of a dictionary of phrases, etc., to recognize one section of information, such as a phrase, for the input information.

However, according to such conventional methods, all phonemes,
A standard pattern such as a syllable and an input cover pattern are matched to calculate the degree of similarity, and candidate syllables i are output in order of similarity.

Therefore, 1. For example, when recognizing monosyllables including syllables, it is necessary to match over 100 types of monosyllable standard patterns and input kabata patterns for each syllable, which takes a lot of processing time. It had become.

In addition, after that, it is necessary to perform dictionary matching processing on all candidate unit sequences output from those with high similarity.
The processing time becomes long, the accuracy of recognizing correct phrases, etc. does not improve, and as a result, the amount of processing required for overall recognition becomes enormous.

<Purpose> The present invention aims to provide a recognition device that eliminates the above-mentioned conventional drawbacks, and improves the accuracy of recognizing a section of information such as a correct phrase, and also improves the accuracy of recognizing a section of information such as a correct phrase. The present invention provides a recognition device that can specify processing according to the type of information to be recognized, and as a result can reduce the amount of processing required for overall recognition.

<Example> Hereinafter, an example in which the recognition device of the present invention recognizes one segment of audio input such as a phrase using more subdivided unit elements such as syllables will be described as an example.

According to an embodiment of the present invention, in a recognition device that recognizes information to be recognized, such as a segment of speech such as a phrase, using N unit elements that are further divided into phonemes, kana, syllables, etc., the recognition target A transition that stores multiple types of different transition matrices that describe transition relations, which are connections between (N+1) characters (unit elements), for character (unit element) strings such as clauses or sentences, for each topic or field. matrix storage means;
A transition matrix specifying means for specifying a desired transition matrix according to the content of the character string to be recognized (type 1 field, etc.) from among the plurality of types of transition matrices stored in the transition matrix storage means, and the transition matrix storage means When the syllable (unit element) lattice is generated based on the transition matrix specified from the number, there is no transition from the i-complementary syllable (unit element) of the previous - syllable (unit element). Groups are excluded from the recognition control, and (when creating a matrix and a candidate column, the transition matrix is referred to for each candidate column, and candidate 7m columns that include a combination of syllable c unit elements that do not transition) are excluded. It is configured to include processing means for performing processing, and to reduce the amount of processing in the next high-order dictionary matching.

First, prior to a detailed explanation of the present invention, a series number with transition 1 indicating a transition relationship, which is a connection relationship between unit elements used in the recognition device of the present invention, will be explained.

In general, when a Japanese sentence is expressed entirely in kana characters, it can be expressed as a string of syllables corresponding to the kana character string.

For example, the phrase ``Earth's'' is made up of four monosyllable unit elements called ``chi゛``kiyu'' and ``u゛''゛゛.The connection relationship between the two syllables (``chiriki'') kyu, from ``kiyu'' to ``u, ``u゛ to ゛ no ゛) can be connected if you look up sentences in all Japanese or two topics in a specific field (transition; the word transition will be used below) There are some syllable pairs that do not come before the syllables of the line. For example, nothing other than 1 "n++, Ittsu" 5 comes before the syllable in the line. Also, paniya' does not come at the beginning of a word, and "hepa (something pronounced as he) does not come at the end of the word.

The first-order transition relationship of the syllables constituting such a bunsetsu is described according to the following equation (1), and a transition matrix M(X, Y) as shown in FIG. 1 is created.

In FIG. 1, the transition matrix M(M, Y) describes the transition from character X to the next character Y in a character string, which is a unit element string, and when there are N unit elements (syllables), ( N+])
It is a matrix of x(N+1), and is stored in a ROM or the like in terms of a node. In addition, column 70 indicates whether each unit element (1 to N) comes in a clause, and row X0 indicates each unit element (1 to N).
Data indicating whether or not N) comes at the end of the knot is written.

For example, Figure 2 shows an example in which the transition of the character string °“red゛” is written in a transition matrix.The elements of the transition matrix can be either 0 (transition not possible) or 1 (transition possible). expressed, 1
It is stored in bits. In addition, in Figure 2, the notation ++
All matrix elements other than +++ are "0'" and their display is omitted.

Next, the creation of the transition matrix will be explained in a little more detail.

First, when creating a transition matrix, the transition matrix memory is initialized to 0'' (CMCX,Y)-0].

Next, the string bar (al, a2 +aB +..., a
l) However, when ■=number of characters in a string, the character transition relationship of the character string A\ is expressed as a transition matrix M(X
, Y). Similarly, the creation of a transition matrix (primary) in which transition relationships are written for all character strings to be recognized is completed.

The concrete transition matrix (first order) M created in this way
An example of (X, Y) is shown in FIG.

As is clear from FIG. 3, for example, the bit position of (X, Y) = (E, <) is "'pi," so from "E" to "
There is a transition to ``, and (X, Y) = (Eh.

Since the bit position of ke) is 0 "', it means that there is no transition from "e" to "ke".

1 is a first-order transition, but a second-order transition, and generally an M-order transition matrix expanded to an M-order, can be similarly created according to the following equation (2).

M-order transition matrix: Mcxl, x2. x3. ...+XM+
, Y) + CN publication) M + 1-dimensional M (a + y + a
+-CM-1) Place・+ ai) −1+ (+ =I
-I+1)-(2) The embodiment of the present invention provides a plurality of transition matrices for each type of recognition target, 7 topics, 2 public areas, etc.
This allows recognition processing to be performed by selecting a specific transition matrix as necessary.

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 4 is a block diagram showing the configuration of an apparatus according to an embodiment of the present invention.

In FIG. 4, reference numeral 1 denotes a transition matrix designation means, and the designation means 1 is connected to a central processing unit (CPU) and is constituted by a selection key provided on an operation surface or a manual selection means by voice. Further, 8 is an input section into which voice information to be recognized is input, 4 is an amplification section, 5 is an acoustic processing section, 61, 6
2. . . , 6 is a transition matrix storage means for storing different types of transition matrices, and 7 is a recognition processing section.

In the above configuration, transition matrix storage means 61, 62,
. . , 6K stores different types of transition matrices created from texts in different fields (for example, science, literature, economics, etc.). If it is related, operate the transition matrix specifying means 1 to select and specify a transition matrix storage means (for example, M + ) created from science-related texts, etc., and store this selection and specification in the memory M1. A recognition processing operation is performed in the recognition processing section 7 using the transition matrix.

Next, the type of information that should be recognized as described in 1. Field C)
A recognition operation using a transition matrix selected and specified according to the following will be explained.

FIG. 5 is a detailed block diagram of the acoustic processing section 5 and the recognition processing section 7 shown in FIG. 4 above.

In FIG. 5, the speech information input to the phrase speech input section 21 is input to the next stage acoustic processing/comparison section 22. This acoustic processing/comparison section 22 is of a conventionally known type except for a processing section using a transition matrix memory 26. For example, the phrase speech signal input to the phrase speech input section 21 is processed by the acoustic processing section 22.
Feature extraction processing is performed for each single syllable, and the feature pattern for each single syllable is temporarily stored in a buffer within the processing unit 22. On the other hand, the storage device 23 stores a standard pattern PH (i=I to N) for each monosyllable, and this standard pattern P1 is sequentially read out and stored in a buffer in the processing/comparison section 22. A matching calculation is performed with the input feature pattern of the input voice.

According to the prior art, this matching calculation process between the standard pattern and the human feature pattern was performed for all standard patterns, but according to the present embodiment, as will be described later, the matching calculation process is performed in the transition matrix memory 26. Based on the information, the matching with the standard pattern of syllables that can be connected to the syllable previously recognized as a candidate (in the first case, a syllable that may come at the beginning) is calculated, and the most similar one is the first candidate. , and the successive approximations are selected as the next candidate,
The result is stored in the candidate syllable memory 24. That is, when generating a syllable lattice, syllable groups that do not transition from any candidate syllable before the -syllable are excluded from recognition comparison.

The transition matrix memory 26 corresponds to one memo 'J (Mi) in the transition matrix storage means 6], 62°, . . . , 6 designated by the transition matrix designation means 1.

The time series of a plurality of x*N syllables stored in the candidate syllable lattice memory 24 is input to a candidate string output section 27 consisting of a candidate string creation section 25 and a transition matrix memory 26, and in the candidate string output section 27, By referring to the contents of the transition matrix memory 26 corresponding to a specific topic field, etc., candidate strings that include syllable transitions that cannot be transitioned are excluded, and only candidate strings that are transitionable are selected.
Reliability: 4.4. >' The candidate string is created in the order of combination, and the dictionary collation unit 29 collates this candidate string with the clauses stored in the dictionary 28, and if they match, the result is output to the clause output unit 30.

Next, syllable recognition processing using the transition matrix M(X, Y) will be explained with reference to a block diagram of candidate syllable creation processing using the transition matrix shown in FIG.

In this embodiment, the resulting candidate syllables are temporarily stored in the candidate syllable lattice buffer 24 in chronological order. Further, the above-mentioned transition matrix information is stored in the memory 26, and the syllable standard pattern is stored in the memory 23.

The recognition results are stored in the candidate syllable lattice 24 as shown in the table below, and when the i-th syllable is to be recognized, the following processing is executed.

However, J(i)' Number of i-th syllable candidates Sl,: j-th syllable ■ Candidate syllable number order, previous syllable candidates X = (si ,, j) j-1 to J (i-1) Number of combinations:
J(i-]) When C1-0, when Sl, j=o, the immediately preceding multiple (J(i-)
1) The sum of the transition matrices is calculated for the candidate syllables, and the syllables whose row m(Y) obtained is O are designated as non-transitionable.

m(Y)-VM(S-Y) ---
−−・−t311′+3+=M(SY)→−M(SY)+・
→i-1, I, i-1,2.

M(Si I, Jci-') + y) In this equation (3), m(Y)-o, excluding the syllable group designated as non-transitionable, performs the next similarity comparison process, and then The candidate syllable of the i syllable is output and written in the candidate syllable lattice 7. However, in the case of 1-1 (syllables of clause class), the syllable group designated as non-transitionable by the 0th row M(0, Y) is excluded and the similarity comparison process is performed.

By repeating the above steps, the creation of the candidate syllable lattice of the -phrasal speech is completed.

Now, when "Kokuminwa" is input as the -syllable speech, the acoustic processing unit 22 extracts features for each syllable, and the feature pattern for each syllable is stored in the input pattern time series buffer 31. Next, the process moves to candidate syllable creation using the transition matrix, and first, the feature pattern of the first syllable is read into the input cover turn buffer 32, and then step n
3, the rows of the transition matrix are specified using the previous candidate syllable group according to equation (3). In the first case, M(0, Y) in the 0th line is specified in step n4, its contents are temporarily stored in the buffer 33, and the occurring syllable is specified in step n5.

Next, proceeding to step n6, the input cover turn buffer 32
The characteristic pattern of the first syllable x 1 stored in
A similarity comparison is made between the standard patterns designated as occurring syllables by the standard pattern internal buffer 33 stored in the standard pattern buffer 33 and sequentially read out to the standard pattern buffer 34 (step n7), and based on the result, candidates are selected. The syllables are output (step n8), and the result is a candidate syllable lattice 24.
written in. In this embodiment, "KO"', "+G OII, and I"BO" are stored as first syllable candidates.

Next, the process returns to step n2, where the second syllable feature pattern x2 is input to the buffer 32, and the process proceeds to step n3, where +IK is added based on the first candidate syllable of the candidate syllable lattice 24.
O! 1, M(S,, 1 to B+y) of each row corresponding to l "GO°゛, BO'" is specified, and in step n4, the sum (OR) of the transition matrices is created and the result is stored in the buffer 33. The syllable of step n5 is specified.

Next, the process moves to step n6, and similar steps n6 to
Execute n9 to select the second candidate syllable "'KU'", "G U"
is stored in the memory 24.

By repeating the above operations, the creation of a candidate syllable lattice for one phrase is completed.

Candidate examples are stored in the candidate syllable lattice 24 as described above, but examples for the conventional method and the wooden sword method when no transition matrix is used are shown in the table below for the input voice "Kokumin wa". show.

” - As is clear from the example below, it can be seen that the correct character strings are higher in the candidate string when using the 1-bokuto method.

Although the transition matrix ``-'' is a first-order transition, it can be extended to a second-order transition, or even a general M-order transition, using the same method.

Note that the M-order transition matrix is created according to the above equation (2), and n
Syllable designation from the iJ candidate syllable (up to the M syllable) can be performed using the following equation (4).

That is, the M-order transition matrix M(X, ,X2 ,=・,'XM,
In the case of expansion to Y ), the previous syllable candidate sequence is +x, , x 2. - TSXM) = (Si-M,j
l Si −(M−1), j2 °” Si −1,
jM) jl-1~J(i-M) j2-1-J(i-(M-1)) jM-1-J(+ 1) Number of combinations: J(i-M)・J(i- CM-1))・・
・If J(i-1) (S, ,, -0 when l!<0), the syllable specification is +-M, J 1.1(M-1), j2.・・・＋ Sl
−+,jM, Y)−(i)m(Y)=VM(S
-S.

j, =l~J(i-M) j2-1~J (i-(M-1)) jM-1~J(i-1) This will be done by.

Note that when the order of M is increased, the syllables to be generated become more limited, and the effect becomes larger than 1.

Next, the operation of creating a candidate syllable string using a transition matrix, which is executed by the candidate string output section 27, will be explained with reference to the process block diagram of creating a candidate string using a transition matrix shown in FIG.

Based on the contents of the candidate syllable lattice memory 24 that stores the time series of a plurality of candidate syllables output from the acoustic processing/comparison section 22 shown in FIG. Candidate strings are created in ascending order, and the results are temporarily stored in the candidate syllable string buffer 42. The candidate syllable string stored in the candidate syllable string buffer 42 is converted to the transition matrix M stored in the memory 26 by the transition matrix reference unit 43.
(X.

Y], the determination section 44 determines whether the transition is possible or not using the following equation (5), and only possible candidate strings are sent to the candidate syllable string output buffer 4 via the candidate syllable string writing section 45.
I will remember it in 6.

If the candidate syllable string of the J-th younger brother is bar (al+a2+..., al), where a,: the first syllable number ■ the number of syllables in one column, the transition matrix M(X, Y
) is used to negate a candidate sequence if any one of the following is true.

In this formula (5), candidate syllable strings containing non-transitionable syllable strings in which any one of them is true are excluded, the same judgment is made for the next candidate syllable string, and only transitional candidate syllable strings are output. The data is stored in buffer 46.

Now, when "Kokuminwa" is input as the -syllable sound, the candidate syllable lattice memory 4 is processed by the acoustic processing/comparison unit 2.
Candidate syllables as shown in the following table are stored in chronological order.

Based on the syllable lattice stored in the memory 24, candidate columns are created in descending order of reliability, and transition matrices: MCX, Y
), only transitionable candidate strings are output, and in this example, candidate syllable strings are output as follows.

According to the conventional method that does not refer to the transition matrix, rGOKUI) INWAJ would be output as the candidate string with the highest reliability, but according to this method, the syllable transition of this candidate string, for example from 'KU' to 'PI ” is determined to be transition-invalid using the transition matrix: M(X, Y), and is excluded from subsequent dictionary matching processing.

Although the transition matrix described above is a first-order transition, it can be extended to a second-order transition or even a general M-order transition using the same method.

Note that the M-order transition matrix can be created according to the above equation (2), and the candidate syllable string can be negated using the following equation (6).

That is, the M-order transition matrix 'M(XI + X2 +'''+
xMI Y ), the j-th candidate column is A j
-(a 1 + a 2 +...

al) then M(a; 1.aH(Ml), ,, al) = OC1 river ~
Negation is performed when any one of I + 1 )-113+ (a i -0 when l≦0.l>1) holds true.

Note that if the degree of M is increased, the candidate syllable string becomes more limited, and the effect becomes greater.

As in Chichifu, when creating a candidate string, the matrix M is referred to for each candidate string, and candidate strings that include combinations of syllables that do not transition are excluded.

The recognition target of the recognition device described above is not limited to phrases, but may also be syllables or single-word sentences, and the subdivided units are not limited to syllables, but may also be phonemes or words.

Also, character strings such as alphabets or FORTRAN
It may be a character string of a programming language such as a language.

In general, the present invention can be applied to any character string in which there is a transition relationship between subdivided units constituting a recognition target word.

As described above, according to the present invention, it is possible to extract correct candidate sequences with high accuracy, so the accuracy of recognizing correct phrases, etc. is increased, and as a result, the amount of processing such as high-level dictionary matching can be reduced. types of information that can be reduced and should be recognized;
Depending on the content, 9 topics, 2 public fields, etc., it is possible to arbitrarily select and use transition matrices for each topic, 9 fields, etc., as necessary each time, which increases the effectiveness of recognition processing using transition matrices. It is possible to do so.

In addition, in the present invention, transition matrices of different types with the same degree, such as those created for sentences and clauses for each topic;
, a new transition matrix; M (M=MiUMj) can be easily created by summing and composing them from .

[Brief explanation of the drawing]

Figure 1 shows a linear transition matrix, Figure 2 shows an example of a transition matrix in which character string transitions are written, Figure 3 shows an example of a transition matrix for bunsetsu character strings, and Figure 4 shows an example of a transition matrix in which character string transitions are written. A block diagram showing the configuration of an embodiment of a recognition device implementing the present invention, FIG. 5 is a detailed block diagram of a recognition processing unit using a transition matrix, and FIG. 6 is a processing flow for creating candidate syllables using a transition matrix. 7 are processing block diagrams for creating candidate columns using a transition matrix. 1...Transition matrix designation means, 2...Central processing unit (CP)
U), 61 , 62 , . . . , 6K . . . transition matrix storage means, 7 . . . recognition processing unit. Master (Treasury) Figure 1 End of story Beginning of story ■ 0″″″ Base, Figure 3

Claims (1)

[Scope of Claims] 1. In a recognition device that recognizes one section of information to be recognized using N unit elements that are further subdivided, the information between (N+1) unit elements for a predetermined unit element string to be recognized is Transition matrix storage means for storing a plurality of different transition matrices that describe transition relationships that are connection relationships; and transition matrix designation means for specifying a predetermined transition matrix among the plurality of different transition matrices stored in the transition matrix storage means. and processing means for performing recognition processing based on the transition matrix designated by the transition matrix designation means. 2. The recognition device according to claim 1, wherein the information to be recognized in one section is speech information in units of words or phrases, and the unit element string is character strings in units of words or phrases. 3. The recognition device according to claim 1, wherein the plurality of different transition matrices are a plurality of transition matrices created from texts in different fields.