JP2000330586A - Method and device for recognizing speech - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech recognition device permitting to correct a partial word string or a partial character string of the recognition result obtained over the whole input speech by arbitrarily specifying them. SOLUTION: This speech recognition device comprises a speech recognition part 10 for recognizing input speeches in linguistic units and generating a graph expressing word strings by the arcs corresponding to the linguistic units, a segmentation part 23 for specifying an arbitrary time segment, and a language processing part 20 for generating plural recognition results concerning the arbitrary time segment specified by the segmentation part 23 in the graph generated by the speech recognition part 10.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for recognizing continuously generated speech.

[0002]

2. Description of the Related Art Japanese Patent Application Laid-Open No. 9-28
Japanese Unexamined Patent Publication No. 1989 discloses a speech recognition apparatus that can omit useless collation in linguistic evaluation processing and can perform recognition processing in a realistic time. FIG. 9 shows a schematic configuration of the voice recognition device. As shown in FIG. 9, the speech recognition device of the above publication includes a phoneme recognition unit 110, a phoneme model storage unit 111, a language processing unit 120, a candidate storage unit 121,
Result storage unit 122, dictionary storage unit 130, linguistic information storage unit 130, syntax rule storage unit 140, linguistic information storage unit 150
Consists of

The phoneme recognition unit 110 divides an input speech into phoneme units (segmentation), recognizes each state of the divided section with reference to the phoneme model storage unit 111, and obtains a phoneme recognition result as a phoneme recognition result. A graph representation of a unit (a phonological graph expressing the phonological structure of a word through a network) is output. This phoneme recognition unit 110
Is a network model connected by a plurality of nodes with the start of the utterance as the start node and the end of the utterance as the end node, and the state of each node section is represented by a phonological symbol or a pseudo-phonological symbol. And an arc representing the transition between the states is provided. Each arc corresponds to a recognized phoneme unit, and a phoneme matching score and a monogram score for the phoneme unit are given as recognition scores. The recognition score indicates a measure of the likelihood of the recognized phoneme candidate in the node section or an arc transition probability. Each node is given the best score from that node to the end node (end of speech).

The language processing unit 120 includes a dictionary storage unit 130 that stores information on recognizable words based on the phoneme graph output from the phoneme recognition unit 110, and a syntax rule storage that describes a sentence that can be accepted from the part of speech. The final recognition is performed using the unit 140 and the linguistic information storage unit 150 storing statistical linguistic information.

Next, a specific processing flow of phoneme recognition in the above speech recognition apparatus will be described with reference to FIG.

When a speech is input, the phoneme recognition unit 110 performs phoneme recognition on the input speech to create a phoneme graph (step S101). This phonological graph is supplied to the language processing unit 120, and the language processing unit 120 executes the following language processing in steps S102 to S109.

First, a candidate storage unit for storing a candidate being processed is initialized to a state in which only one initial candidate is stored (step S102). Next, after confirming that the candidate storage unit is not empty, the candidate with the highest score is extracted from the candidate storage unit (steps S103 and S10).
4). However, initially, the initial candidates prepared in step S102 are extracted.

When the candidate with the highest score is retrieved in step S104, it is then determined whether or not the retrieved candidate reaches the end of the phoneme graph (step S105). If the collation has reached the end of the phonological graph and is established as a sentence, the candidate is moved to the result storage unit (step S106). Enter processing.

After the candidates are moved to the result storage unit in step S106, it is determined whether the number of candidates moved to the result storage unit is sufficient (step S107).
If it is insufficient, the process returns to step S103, and if it is enough, the candidate moved to the result storage unit is output as a recognition result (step S108). The recognition result output in step S108 is also performed when the candidate storage unit becomes empty in step S103.

In the phoneme recognition processing described above, the language collation processing proceeds from the node where the processing is completed to the subsequent node. Specifically, processing is performed in which a node that can be accepted is selected as a new candidate using the dictionary storage unit 130 and the syntax rule storage unit 140, and the candidate is subjected to linguistic evaluation to give a score. . Here, an evaluation score is obtained with reference to the linguistic information storage unit 150. The evaluation score includes a prediction score on the phoneme graph.

In the conventional speech recognition apparatus described above, if the condition that the predicted score does not fall below the actual score is satisfied, a better evaluation score is obtained in the order in which the recognition results are obtained. That is, for a phoneme graph obtained from a certain input voice, recognition results from the start to the end of the phoneme graph can be obtained in the order of scores. For example, if the number of candidates to be stored in the candidate storage unit and the number of recognition results to be stored in the recognition result storage unit are not limited, all recognition results that connect from the beginning to the end of the phoneme graph are obtained in the order of scores. Can be.

[0012]

If there is an error in the partial word string or partial character string in the first place recognition result obtained for the entire input speech, the recognition process can be performed efficiently if only that part can be corrected. it can. However, the above-described conventional speech recognition device obtains a plurality of recognition result candidates for the entire input speech and selects candidates from the candidates in descending order of the evaluation score. The partial word string or partial character string in the recognition result could not be corrected.

An object of the present invention is to provide a speech recognition method and a speech recognition apparatus capable of arbitrarily designating and correcting a partial word string or a partial character string in a recognition result obtained for an entire input speech. It is in.

[0014]

In order to achieve the above object, a speech recognition apparatus according to the present invention recognizes an input speech in a linguistic unit, and a graph in which a word string is represented by an arc corresponding to the linguistic unit. , A section specifying means for specifying an arbitrary time section, and a plurality of recognition results for an arbitrary time section specified by the section specifying means in a graph generated by the voice recognizing means. And a language processing means for generating

The speech recognition method according to the present invention includes a speech recognition step of recognizing an input speech in a linguistic unit and generating a graph in which a word string is represented by an arc corresponding to the linguistic unit; A language processing step of generating a plurality of recognition results for an arbitrarily designated time section in the graph generated in the step.

(Operation) In the present invention as described above, a plurality of recognition results are generated for an arbitrary time section of the graph obtained from the input speech, and therefore, among the first-order recognition results obtained for the entire input speech. If there is an error in the partial word string or partial character string, a plurality of recognition results can be obtained for that part. The user can correct an erroneous part by arbitrarily selecting a valid result from these recognition results.

[0017]

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

FIG. 1 shows an embodiment of the speech recognition apparatus of the present invention. The speech recognition device according to the present embodiment includes a speech recognition unit 10, a standard pattern storage unit 11, a language processing unit 20, and a candidate storage unit 2.
1, a result storage unit 22, a section designation unit 23, a prediction score calculation unit 24, and a linguistic information storage unit 30.

The speech recognition unit 10 analyzes the standard pattern stored in the standard pattern storage unit 11 and the dictionary information stored in the linguistic information storage unit 30 for the parameter vector sequence obtained by analyzing the input speech. Using the linguistic information, generate a graph (hereinafter referred to as a word graph) in linguistic units. For analysis of the input speech, for example, a filter bank, a Fourier transform, a linear prediction coefficient type analyzer, or the like is used. As a linguistic unit of the word graph, phonemes, syllables, words, and the like can be used.

In the word graph, linguistic units are represented by arcs. Each arc has an acoustic score (acoustic score) representing the proximity of the linguistic unit between the corresponding part of the analyzed input speech and the standard pattern. Showing a measure of certainty)
Is given. The word graph has one start node and one end node, which correspond to the start and end of the input speech, respectively. The nodes of the word graph have information corresponding to the time position in the input voice.

The standard pattern storage 11 stores voices analyzed in advance. For example, speech is stored in phoneme units, phoneme units in consideration of the surrounding environment, syllable units, and word units.

The language processing section 20 is based on the word graph provided from the speech recognition section 10,
The evaluation score of the recognition result candidate is calculated using the linguistic information stored in 0, the acoustic score given to the word graph, and the prediction score obtained for each node of the word graph by the prediction score calculation unit 24. By storing the recognition result candidates in the language processing in the candidate storage unit 21 and storing the recognition result candidates for which the language processing has been completed in the result storage unit 22 as the recognition results, it is possible to correspond to the section specified by the section specifying unit 23. Recognition results are obtained in the order of evaluation scores for the sections in the word graph to be executed.

The candidate storage unit 21 stores recognition result candidates obtained when the word graph is being processed in the language processing unit 20 in the order of good evaluation score. Here, the recognition result candidates include at least a candidate evaluation score, a score in the middle of the candidate, a node number processed recently, and 0 or 1.
Is set as the minimum information.

The result storage unit 22 stores the recognition results obtained by processing the word graph in the language processing unit 20 in the order in which the recognition results are obtained. The section designation unit 23 gives a time section for which a recognition result is to be obtained. The user can directly specify the time in the input voice or the partial word string or partial character string in the first-order recognition result (the most reliable recognition result) for the entire input using the section specifying unit 23. This section designation unit 23
It can also be realized by using key input means or the like.

When a partial word string in the first place recognition result is specified by the section specifying unit 23, the time recorded at the start node of the first arc of the path of the corresponding word graph is set to the start time of the section. And the time recorded in the terminal node of the last arc is set as the end time of the section. If a substring in the first place recognition result is specified, for the first arc in the corresponding word graph path, the correspondence between the linguistic unit assigned to the arc and the specified character string, and the start of the arc An estimated value of the start time of the section is obtained from the time recorded in each of the node and the end node. Similarly, the estimated value of the end time of the section is obtained for the last arc.

The prediction score calculation unit 24 includes a language processing unit 20
For each node of the word graph to be processed in the above, using the linguistic information stored in the linguistic information storage unit 30 and the acoustic score assigned to the word graph, calculation is performed from the node toward the start node. The backward prediction score calculated and the forward prediction score calculated from the node toward the end node are provided. Note that these prediction scores may be calculated only for necessary nodes, or may be calculated for all nodes. Further, depending on the linguistic information, the prediction score can be calculated at high speed by using the dynamic programming.

The linguistic information storage unit 30 stores dictionary information and linguistic information used by the speech recognition unit 10 and the linguistic processing unit 20. The dictionary information defines a linguistic unit of the word graph. Although there is no need for linguistic information, a more accurate recognition result candidate can be obtained by using a constraint on the linguistic unit. As the linguistic information, for example, for a word, a word n-gram, an inter-word connection availability determination table, a word part-of-speech connectivity availability determination table, or the like can be used.

Next, the overall operation of the speech recognition apparatus of the present embodiment will be described in detail with reference to the flowchart of FIG.

When a voice is input, the voice recognition unit 10 performs a phoneme recognition process on the input voice to create a word graph (step S1). The word graph is
It consists of a node corresponding to the time position in the input speech and an arc connecting the node, and the arc records a start node, an end node, a corresponding linguistic unit, and an acoustic score. For example, "Computer Speec
h and Language (1997) 11, pp43-72 ".

After the word graph is created, the prediction score calculation unit 24 then calculates, for each node of the word graph, an optimal score of a path from the start node to the node (hereinafter, a backward prediction score). Calculation is performed (step S2). The score of the path is a value obtained by weighting and summing the acoustic score given to the arc constituting the path, the chain of linguistic units of the arc, and the language score given by the linguistic information storage unit 30. When the linguistic information provided by the linguistic information storage unit 30 is determined by two or less linguistic units, this calculation is performed by using dynamic programming in the direction from the start node to the end node to reduce the number of nodes in the word graph. It can be done in proportional time. When the backward prediction score is calculated, subsequently, the prediction score calculation unit 24
Calculates the optimal score of the path from the node to the end node (hereinafter, the forward prediction score) for each node of the word graph (step S3). Also in this calculation, when the linguistic information provided by the linguistic information storage unit 30 is determined by two or less linguistic units, the processing is performed from the end node to the start node using the dynamic programming method, so that the word graph This can be done in a time proportional to the number of nodes.

Next, the language processing section 20 sets the section designating section 23
Based on a start time, an end time, and a predetermined error representing an allowable range of the time section in which a candidate for a speech recognition result specified by the following is to be obtained. , And a node group within the section are obtained (step S4). The section start node group is all nodes existing between “start time−error” and “start time + error” among the nodes in the word graph. The section end node group is calculated from “end time−error” to “end time + error” among the nodes in the word graph.
Are all nodes that exist between The intra-section node group is all nodes existing between “start time−error” and “end time + error” among the nodes in the word graph. If either the section start node group or the section end node group becomes empty in step S4, the language processing unit 20 indicates that a recognition result cannot be obtained, and ends the processing.

When the section start node group, the section end node group, and the intra-section node group are obtained, the language processing section 20
Stores in the candidate storage unit 21 the recognition result candidates obtained by connecting all the nodes of the section start node group to the nodes of the intra-section node group following the node (step S5). The recognition result candidates recorded in the candidate storage unit 21 include each information of a node sequence (limited to the intra-section node group), an evaluation score, an intermediate score, and a processing end flag in the already processed word graph. . For example, as the intermediate score, a backward prediction score of the section start node, an acoustic score (described in a word graph) of an arc connecting nodes of a node group in the section following the section start node, and a language score obtained by the language information storage unit 30 are used. Each node has a value obtained by weighting and summing, and as an evaluation score, a value obtained by weighting and summing the forward prediction scores of the nodes in the node group in the section connected to the obtained intermediate score, as a processed node sequence The candidate storage unit 21 has a section start end node and a node of a node group in the section connected to the section start node, and has a recognition end candidate having 0 as a processing end flag. Note that a penalty (score in the negative direction) proportional to the difference between the time recorded at the section start node and the start time can be added to the midway score.

After the recognition result candidates are stored in the candidate storage unit 21, the language processing unit 20 checks whether the candidate storage unit 21 is empty (step S6). If not empty,
The recognition result candidate having the best evaluation score is extracted from the candidate storage unit 21 and the information of the candidate is stored in the candidate storage unit 21.
It is deleted from within (step S7). Then, it is checked whether the extracted recognition result candidate has 1 as a processing end flag (step S8).

If the processing in step S8 does not have 1 as the processing end flag, the language processing section 20 proceeds with the following recognition result candidate creation processing for the extracted recognition result candidates (step S9).

When the most advanced node in the node sequence stored in the extracted recognition result candidate is included in the section end node group, the recognition result candidate is set to a new recognition with the processing end flag set to 1. It is stored in the candidate storage unit 21 as a result candidate. In this case, the evaluation score may be left as it is, or a penalty proportional to the difference between the time recorded in the section end node and the end time may be added.

Contrary to the above, when the most processed node is not included in the section end node group, the recognition result candidates obtained by connecting the nodes of the node group in the section following the node are stored in the candidate storage section 21. To memorize. In this case, the intermediate score is the intermediate score stored in the recognition result candidate, and the acoustic score connecting the nodes of the node group within the section following the most processed node in the node sequence stored in the recognition result candidate. , And a language score. The evaluation score is obtained by weighting the newly obtained midway score and the forward prediction score of the connected node. In this case, the processed node sequence, that is, the node sequence obtained by adding the nodes of the intra-section node group connected to the node sequence stored in the recognition result candidate is recorded, and 0 is recorded as the processing end flag. The obtained recognition result candidates are stored in the candidate storage unit 21.

In the process of step S8, when the extracted recognition result candidate has 1 as the processing end flag, the language processing unit 20 stores the extracted node sequence of the recognition result candidate in the result storage unit 22 as the recognition result. (Step S10). Then, it is determined whether or not the number of recognition results stored in the result storage unit 22 has exceeded a predetermined number (step S11).

In step S11, when the number of obtained recognition results is sufficient, or in step S6 described above.
In the case where the candidate storage unit 21 is empty, the language processing unit 2
0 outputs the recognition result stored in the result storage unit 22 and ends the process (step S12). If a sufficient recognition result has not been obtained, the language processing unit 20 outputs that fact.

In the above speech recognition processing, the forward / backward prediction scores for the nodes of the word graph obtained by the processing up to step S3 do not depend on the section for which the recognition result is to be obtained. When re-acquiring, it is only necessary to start over from step S4.

Normally, one word graph is obtained for one utterance, but it is easy to connect a plurality of word graphs into one word graph. Therefore, in the case of the present embodiment, the speech recognition unit is configured to create a graph for each of the input voices for a plurality of input voices that are continuously input, and to create one graph by connecting these graphs. You may. Specifically, it has a graph storage unit that stores one or more word graphs, and the speech recognition unit is configured to connect the word graphs stored in the graph storage unit. In this case, it is possible to obtain a plurality of different speech recognition results in the order of score for a time section spanning a plurality of input speeches. In this case, if linguistic information is used, a context that spans multiple utterances can be used.

Next, the speech recognition processing in the speech recognition apparatus of the present embodiment will be described with a specific example. In FIG.
4 shows an example of a word graph created by a voice recognition unit of the voice recognition device of the present embodiment. This word graph is obtained from the voice of the sentence "Here is quite cold at night". The start node is indicated by "S", and the end node is indicated by "E". , The other nodes are numbered in chronological order. The word corresponds to the arc. Although omitted from FIG. 3, the arc is given the acoustic score of the word. The word graph is generated so that arcs entering the same node become the same word. Here, the flow of processing for obtaining a plurality of recognition results for the time section of the arc “humor” (arc from node 6 to node 8) of this word graph will be specifically described.

First, a forward / backward prediction score is calculated for each node of the word graph as preprocessing by using a dynamic programming method. Specifically, for example, in the case of a backward prediction score, for the start node k of all arcs entering a certain node j, Sb (node j) = max (a (arc kj) +1 (W
k, Wj) + Sb (node k)) compute node k. Here, Sb (node j) is the backward prediction score of node j, a (arc kj) is the acoustic score of the arc entering node j from node k, Wk is the word of the arc entering node k, l (Wk, Wj) Is two words Wk, Wj
Is the language score for

In the order of node numbers (here, "S" is first,
"E" is the last. ) By performing the above processing,
Sb for the starting node k of all arcs entering node j
(Node k) can be calculated. In this calculation,
Weighting may be performed by multiplying each of the acoustic score and the language score by an appropriate coefficient. The language score is a table relating to two words as shown in FIG. A high-speed search method (for example, a binary search) may be used to search the table at high speed. The obtained prediction score is stored as a table for each node of the graph as shown in FIG. Although partly omitted in FIG. 5, it is actually calculated and stored for all nodes of the word graph.

Next, a section start node group, a section end node group, and an intra-section node group are obtained. For simplicity, here,
An error representing an allowable range of a time section is set to 0. From the word graph of FIG. 3, a node “6” having the same time as the time recorded at the start node of the arc “humor” is obtained as a section start node group, and as a section end node group,
Nodes “8, 9, 10” having the same time as the time recorded in the terminal node are obtained.
Nodes “6, 7, 8, 9, 10” included between the time recorded in the start node and the time recorded in the end node
Is required.

Next, for all the nodes in the section start node group, recognition result candidates are created for the subsequent nodes in the section node group and stored in the candidate storage unit 21.
At this time, the evaluation score of the candidate is calculated. In this evaluation score calculation, one of the nodes of the section start node group is set to the node i
G (candidate i, j) = Sb (node i) + a (arc i)
j) + l (Wi, Wj) S (candidate i, j) = g (candidate i, j) + Sf (node j) Here, g (candidate i, j) is an intermediate score of the recognition result candidate, S (candidate i, j) is an evaluation score of the recognition result candidate, and candidates i, j are a node sequence {node i, node j}.
, Sf (node j) is the forward prediction score of node j. 4 and 5, recognition result candidates as shown in FIG. 6 are stored in the candidate storage unit 21.

In the subsequent processing, since the candidate storage unit 21 is not empty, a recognition result candidate having the best evaluation score is extracted.
If the processing end flag is 1, the node sequence is stored in the result storage unit 22 as a recognition result. Otherwise, a new recognition result candidate is created based on the extracted recognition result candidate and stored in the candidate storage unit. . Specifically, the following processing is performed with the most processed node in the processed node sequence of the extracted recognition result candidate as the node i.

If the node i is a node in the section end node group, the recognition result candidates with the processing end flag set to 1 are stored in the candidate storage unit 21. If the node i is not a node of the section end node group, the following is applied to the node j of the node group within the section: g (candidate + j) = g (candidate) + a (arc ij) +1
(Wi, Wj) S (candidate + j) = g (candidate + j) + Sf (node j) is calculated. Here, “candidate” is the extracted recognition result candidate, and “candidate + j” is a recognition result candidate having a node sequence obtained by adding the node j to the processed node sequence of the extracted recognition result candidate.

The above processing will be specifically described below, taking as an example the case where the recognition result candidates shown in FIG. 6 are stored.

First, a recognition result candidate "candidate 6, 7" having the highest evaluation score is extracted. Since the processing end flag of the extracted recognition result candidate “candidate 6, 7” is not 1, a new recognition result candidate is created. However, since the most advanced node 7 in the node sequence recorded in “candidate 6, 7” is not a node of the section end node group,
A recognition result candidate with the processing end flag set to 1 is not created as a new recognition result candidate. In the word graph of FIG. 3, since only the node 10 in the node group in the section following the node 7 in the node sequence which has undergone the most processing is the node 10, only new recognition result candidates “candidate 6, 7, 10” are created. Is stored. Thereby, the recognition result candidates as shown in FIG. 7 are stored in the candidate storage unit.

Subsequently, the recognition result candidates "candidate 6, 8" having the highest evaluation score from among the recognition result candidates in FIG.
Is taken out. Since the processing end flag of the extracted recognition result candidate “candidate 6, 8” is not 1, a new recognition result candidate is created. In the word graph of FIG. 3, since the most advanced node 8 in the node sequence recorded in “candidate 6, 8” is one of the nodes of the section end node group, “candidate 6” in which the processing end flag is set to 1 , 8 "are created and stored as new recognition result candidates. In this case, since there is no intra-section node group connected to the node 8, no further recognition result candidate is created. Thereby, the recognition result candidates as shown in FIG. 8 are stored in the candidate storage unit.

Subsequently, the recognition result candidate “candidate 6, 8” having the highest evaluation score from among the recognition result candidates in FIG.
Is taken out. Since the processing result flag of the extracted recognition result candidate “candidate 6, 8” is 1, “node 6, 8” is recorded in the result storage unit as the recognition result.

By proceeding with the above-described processing, as a recognition result, “nodes 6 and 8” (score−102), “nodes 6 and 8”
"7, 10" (score-103) and "node 6, 9" (score-105) are obtained in order. These recognition results
Corresponding to the word graph of FIG. 3, "humor", "at night", and "union" respectively. For example, a good user interface can be constructed by presenting in this order and allowing the user to select the correct answer. . As means for allowing the user to select and specify a valid recognition result, for example, key input, mouse input, or the like can be used.

In the speech recognition apparatus of the present embodiment described above,
By obtaining the prediction score of the node of the word graph obtained from the input speech, the recognition result of the section starting and ending at an arbitrary node of the word graph is processed only for the nodes within the time section, Recognition results can be compared and ordered based on the score of the whole path from the beginning to the end of the word graph including the path, and not only the score within the section, but also the overall score and context before and after the section , And a recognition result can be obtained. Therefore, for any section of the input voice,
Different recognition results can be accurately obtained with a small amount of processing.

Further, in the speech recognition apparatus of this embodiment, different recognition results can be obtained for an arbitrary section of one word graph obtained by connecting word graphs obtained from two or more input voices. , And contexts that span multiple utterances. Also in this case, different recognition results can be accurately obtained with a small amount of processing for an arbitrary section.

In the above-described speech recognition apparatus of the present embodiment, the recognition result is stored by using the result storage unit 22, but it may be configured to output the recognition result every time the recognition result is obtained. it can.

[0056]

As described above, according to the present invention,
If there is an error in the partial word string or partial character string in the recognition result obtained for the entire input speech, a plurality of recognition results can be generated for that part, and the user can arbitrarily select any of these recognition results Since the correction can be performed by selecting the result, a correct recognition result can be efficiently obtained with a small amount of processing.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of a speech recognition device of the present invention.

FIG. 2 is a flowchart showing an overall operation flow of the speech recognition apparatus shown in FIG. 1;

FIG. 3 is a diagram illustrating an example of a word graph.

FIG. 4 is a diagram showing an example of a language score.

FIG. 5 is a diagram illustrating an example of a prediction score.

FIG. 6 is a diagram illustrating an example of recognition result candidates stored in a candidate storage unit.

FIG. 7 is a diagram illustrating an example of a recognition result candidate stored in a candidate storage unit.

FIG. 8 is a diagram illustrating an example of a recognition result candidate stored in a candidate storage unit.

FIG. 9 is a block diagram showing a schematic configuration of a speech recognition device disclosed in Japanese Patent Application Laid-Open No. 9-281989.

FIG. 10 is a flowchart showing a specific processing flow of phoneme recognition in the speech recognition device shown in FIG. 9;

[Explanation of symbols]

 Reference Signs List 10 voice recognition unit 11 standard pattern storage unit 20 language processing unit 21 candidate storage unit 22 result storage unit 23 section designation unit 24 prediction score calculation unit 30 language information storage unit

Claims (16)

[Claims] 1. A speech recognition means for recognizing an input speech in a linguistic unit and generating a graph in which a word string is expressed by an arc corresponding to the linguistic unit, and a section designation for designating an arbitrary time section And a language processing means for generating a plurality of recognition results for an arbitrary time section designated by the section designation means in the graph generated by the speech recognition means. 2. The speech recognition device according to claim 1, wherein the speech recognition means generates a graph in which each arc is provided with an acoustic score indicating at least a measure of acoustic certainty, and wherein the language processing means includes: In the graph generated by the speech recognition means, word strings of all arcs present in an arbitrary time section designated by the section designation means are set as recognition result candidates, and for each of the recognition result candidates, at least the sound A speech recognition apparatus characterized in that an evaluation score based on a score is obtained, and recognition results are obtained in order from a recognition result candidate having the highest evaluation score. 3. The speech recognition apparatus according to claim 2, wherein the language processing means uses, for each recognition result candidate, an acoustic score of the entire path from the start to the end of the entire graph including the recognition result candidate. A speech recognition device configured to obtain an evaluation score by using the speech recognition device. 4. The speech recognition device according to claim 2, wherein the linguistic information storage means stores predetermined linguistic information in advance, and the linguistic information stored in the linguistic information storage means is generated by the speech recognition means. Based on the acoustic score given to each arc in the graph, for each node located at the boundary of each arc in the graph, the optimal score of the path from the beginning of the graph to the node and the optimal score of the graph from the node. A predictive score calculating unit for obtaining a predictive score by respectively obtaining an optimal score of the path leading to the end, wherein the language processing unit includes an acoustic score assigned to each arc and a language stored in the linguistic information storage unit. Speech recognition characterized by calculating an evaluation score based on information and a prediction score calculated by said prediction score calculation means. apparatus. 5. The speech recognition device according to claim 1, wherein the language processing means includes a start time and an end time of an arbitrary time section designated by the section designation means. A speech recognition device, wherein a predetermined allowable range is set, and a plurality of recognition results are obtained for a section in which the allowable range is set. 6. The speech recognition apparatus according to claim 1, wherein the section designation means designates a partial word string in a first place recognition result obtained for the entire input speech. A speech recognition device, comprising: 7. The speech recognition device according to claim 1, wherein the section designation means designates a partial character string in a first place recognition result obtained for the entire input speech. A speech recognition device, comprising: 8. The speech recognition apparatus according to claim 1, wherein the speech recognition unit creates a graph for each of the plurality of input speeches that are continuously input. A speech recognition device configured to create one graph by connecting the graphs. 9. A speech recognition step of recognizing an input speech in a linguistic unit and generating a graph in which a word string is expressed by an arc corresponding to the linguistic unit; A language processing step of generating a plurality of recognition results for an arbitrarily designated time section. 10. The speech recognition method according to claim 9, wherein the speech recognition step is a step of generating a graph in which each arc is given an acoustic score indicating at least a measure of acoustic likelihood; The processing step sets word strings of all arcs present in an arbitrarily designated time interval in the graph generated by the voice recognition step as recognition result candidates, and for each of the recognition result candidates, at least the acoustic score A speech recognition method comprising: obtaining an evaluation score based on the evaluation score; and obtaining a recognition result in order from a recognition result candidate having a good evaluation score. 11. The speech recognition method according to claim 10, wherein the evaluation score in the language processing step is calculated for each of the recognition result candidates by using the sound of the entire path from the beginning to the end of the entire graph including the recognition result candidates. A speech recognition method characterized in that it is obtained using a score. 12. The voice recognition method according to claim 10, wherein the graph is generated based on an acoustic score given to each arc of the graph generated in the voice recognition step and predetermined language information prepared in advance. For each of the nodes located at the boundaries of each of the arcs, a prediction score is obtained by obtaining an optimal score of a path from the beginning of the graph to the node and an optimal score of a path from the node to the end of the graph. The method further includes a calculation step, wherein the calculation of the evaluation score in the language processing step is performed based on the acoustic score given to each arc, the linguistic information, and the prediction score calculated in the prediction score calculation step. Voice recognition method to be used. 13. The method according to claim 9, wherein
In the speech recognition method described in the paragraph, the language processing step includes a process of setting a predetermined allowable range for each of a start time and an end time of an arbitrarily designated time interval. A speech recognition method, comprising the step of obtaining a plurality of recognition results. 14. The method according to claim 9, wherein
The speech recognition method described in the paragraph, further comprising the step of designating an arbitrary time section, the step of designating a partial word string in a first-order recognition result obtained for the entire input speech. 15. The method according to claim 9, wherein:
The speech recognition method described in the paragraph, further comprising a step of designating a partial character string in a first place recognition result obtained for the entire input speech, as a step of designating an arbitrary time section. 16. The method according to claim 9, wherein:
In the voice recognition method described in the paragraph, in the voice recognition step, for a plurality of input voices that are continuously input, a graph is created for each input voice, and these graphs are connected to create one graph. A speech recognition method characterized by the following.