JP6222970B2 - Speech recognition apparatus and speech recognition result determination method - Google Patents

Description

  The present invention relates to a speech recognition apparatus that performs speech recognition processing and a speech recognition result determination method used in the apparatus.

  Performing text input by speech input using speech recognition technology (hereinafter referred to as “speech text input”) is widely performed. In recent years, with the expansion of applications, higher speed and higher accuracy of speech text input are required.

  The technique described in Patent Document 1 (hereinafter referred to as “prior art”) divides a sentence into words and sequentially uses a two-level language model and an acoustic model to sequentially obtain speech recognition results from the beginning of the sentence (from the front) in units of words. Determine. By using such a conventional technology, it is possible to perform highly accurate speech text input at a higher speed than when speech recognition results are determined in sentence units using a one-stage language model and acoustic model. It becomes.

Japanese Patent Laid-Open No. 2003-140685

  However, the prior art has a problem that it is difficult to realize high-accuracy and high-speed speech text input for relatively long words.

  The reason is as follows. Long words not only take time from the start to the completion of the utterance, but also take time from the start to the completion of the speech recognition process. Therefore, when a relatively long word is to be input, it takes a relatively long time from the start of utterance until the speech recognition result is confirmed.

  An object of the present invention is to realize high-accuracy and high-speed speech text input even for relatively long words.

The speech recognition apparatus according to the first aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and a speech portion that is already input in the speech. A speech recognition processing unit that performs speech recognition and uses the speech recognition result as a provisional speech recognition result, and an exclusive section in which the provisional speech recognition result is unique among the plurality of text arrays ahead when a exclusive text array with, possess a deterministic processing unit the exclusive text sequences and the speech recognition result of confirmation for the voice, and the voice recognition processing unit repeats the speech recognition in a predetermined cycle performed The confirmation processing unit is forward-matching with all of the predetermined number of the other provisional speech recognition results obtained in the most recent consecutive manner, and the provisional speech recognition result is the exclusive text array. Condition to become, the exclusive text sequences and the speech recognition result of the confirmation, the predetermined number is two.
The speech recognition apparatus according to the second aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and a speech portion that is already input in the speech. A speech recognition processing unit that performs speech recognition and uses the speech recognition result as a provisional speech recognition result, and an exclusive section in which the provisional speech recognition result is unique among the plurality of text arrays ahead A determination processing unit that uses the exclusive text array as a final speech recognition result for the speech, and the speech recognition processing unit repeatedly performs the speech recognition in a predetermined cycle. The confirmation processing unit is forward-matching with all of the predetermined number of the other provisional speech recognition results obtained in the most recent consecutive manner, and the provisional speech recognition result is the exclusive text array. On the condition that the exclusive text array is the confirmed speech recognition result, the text array is a syllable text array, and the confirmation processing unit further includes the exclusive text array and the front in the predetermined number or more of syllables. Text arrays that do not match are not subject to the above conditions .
The speech recognition apparatus according to the third aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and a speech portion that is already input in the speech. A speech recognition processing unit that performs speech recognition and uses the speech recognition result as a provisional speech recognition result, and an exclusive section in which the provisional speech recognition result is unique among the plurality of text arrays ahead A determination processing unit that uses the exclusive text array as a final speech recognition result for the speech, and the speech recognition processing unit repeatedly performs the speech recognition in a predetermined cycle. The confirmation processing unit is forward-matching with all of the predetermined number of the other provisional speech recognition results obtained in the most recent consecutive manner, and the provisional speech recognition result is the exclusive text array. The exclusive text array is the final speech recognition result, the exclusive text array is a syllable text array in which a marker indicating the end position of the exclusive section is inserted, and the plurality of text arrays Among them, the text array other than the exclusive text array is a syllable text array in which a marker indicating the end position of the text array is inserted, and the confirmation processing unit inserts the provisional speech recognition result into the marker inserted. If not, the provisional speech recognition result is not subject to the judgment of the condition .
A speech recognition apparatus according to a fourth aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and a speech portion that is already input in the speech. A speech recognition processing unit that performs speech recognition and uses the speech recognition result as a provisional speech recognition result, and an exclusive section in which the provisional speech recognition result is unique among the plurality of text arrays ahead A speech recognition apparatus having the exclusive text array as a final speech recognition result for the speech, wherein the speech recognition processing unit performs the speech recognition on a predetermined basis. The determination processing unit repeats the provisional speech recognition result in the exclusive text array, and the predetermined number of other provisional speech recognitions obtained continuously in the latest. The exclusive text array is used as the confirmed speech recognition result on the condition that all of the results are in front match, and the speech recognition apparatus sorts the plurality of text arrays described in the dictionary in the order of front match. And a marker insertion unit that inserts a marker into the exclusive text array by performing a process of determining a front matching range for each of the adjacent text array pairs .

The speech recognition result determination method according to the fifth aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and speech that has already been input among the speech. A speech recognition result determination method in a speech recognition apparatus, comprising: a speech recognition processing unit that performs speech recognition on a portion and uses the speech recognition result as a provisional speech recognition result, wherein the provisional speech recognition result Determining whether or not the exclusive text array has an exclusive section that is unique among the plurality of text arrays in front and the provisional speech recognition result is the exclusive text array. It possesses a step of the text sequence and the speech recognition result of confirmation for the voice, and the voice recognition processing unit performs repeatedly the voice recognition at a predetermined cycle, the probability The step of making the speech recognition result of the above-mentioned is that the provisional speech recognition result is the exclusive text array, and a forward coincidence with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. The exclusive text array is the definitive speech recognition result, and the predetermined number is 2 .
A speech recognition result determination method according to a sixth aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and speech that has already been input among the speech. A speech recognition result determination method in a speech recognition apparatus, comprising: a speech recognition processing unit that performs speech recognition on a portion and uses the speech recognition result as a provisional speech recognition result, wherein the provisional speech recognition result Determining whether or not the exclusive text array has an exclusive section that is unique among the plurality of text arrays in front and the provisional speech recognition result is the exclusive text array. And making the text arrangement a final voice recognition result for the voice, wherein the voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle, and The step of making the speech recognition result of the above-mentioned is that the provisional speech recognition result is the exclusive text array, and a forward coincidence with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. The exclusive text array is the definitive speech recognition result, the text array is a syllable text array, and the definite speech recognition result is the exclusive text array with the predetermined number or more of syllables. A text array that does not coincide with the text array is excluded from the judgment of the condition.
The speech recognition result determination method according to the seventh aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and speech that has already been input among the speeches. A speech recognition result determination method in a speech recognition apparatus, comprising: a speech recognition processing unit that performs speech recognition on a portion and uses the speech recognition result as a provisional speech recognition result, wherein the provisional speech recognition result Determining whether or not the exclusive text array has an exclusive section that is unique among the plurality of text arrays in front and the provisional speech recognition result is the exclusive text array. And making the text arrangement a final voice recognition result for the voice, wherein the voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle, and The step of making the speech recognition result of the above-mentioned is that the provisional speech recognition result is the exclusive text array, and a forward coincidence with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. On the condition that the exclusive text array is the final speech recognition result, the exclusive text array is a syllable text array in which a marker indicating the end position of the exclusive section is inserted, and the plurality of text arrays Of these, the text array other than the exclusive text array is a syllable text array in which a marker indicating the end position of the text array is inserted, and the step of setting the final speech recognition result includes the provisional speech recognition result, When the marker is not inserted, the provisional speech recognition result is excluded from the judgment of the condition.
A speech recognition result determination method according to an eighth aspect of the present invention includes a speech recognition database that stores a dictionary describing a plurality of text arrays, a speech input unit that inputs speech, and speech that has already been input among the speech. A speech recognition result determination method in a speech recognition apparatus, comprising: a speech recognition processing unit that performs speech recognition on a portion and uses the speech recognition result as a provisional speech recognition result, wherein the provisional speech recognition result Determining whether or not the exclusive text array has an exclusive section that is unique among the plurality of text arrays in front and the provisional speech recognition result is the exclusive text array. And making the text arrangement a final voice recognition result for the voice, wherein the voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle, and The step of making the speech recognition result of the above-mentioned is that the provisional speech recognition result is the exclusive text array, and a forward coincidence with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. The exclusive text array as the confirmed speech recognition result, and the speech recognition result determination method sorts a plurality of text arrays described in the dictionary in the order of front matching, The method further includes the step of inserting a marker into the exclusive text array by performing a process of determining a front matching range for each of the text array pairs.

  According to the present invention, high-accuracy and high-speed speech text input can be realized even for relatively long words.

The block diagram which shows an example of a structure of the speech recognition apparatus which concerns on one embodiment of this invention. Flow chart showing an example of marker insertion operation in the present embodiment The figure which shows an example of the registration word group before the marker insertion operation | movement in this Embodiment The figure which shows an example of the syllable matrix in this Embodiment The figure which shows an example of the matching character string list | wrist in this Embodiment The figure which shows an example of the registration word group after the marker insertion operation | movement in this Embodiment Flow chart showing an example of voice recognition operation in the present embodiment The figure explaining an example of the time required until the decision result acquisition in this Embodiment The figure which shows an example of the mode of a result decision in this Embodiment The figure which shows the other example of the mode of a result decision in this Embodiment.

  Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

  First, the configuration of the speech recognition apparatus according to this embodiment will be described.

  FIG. 1 is a block diagram showing an example of the configuration of a speech recognition apparatus according to an embodiment of the present invention. The voice recognition device according to the present embodiment is, for example, a mobile phone.

  In FIG. 1, a speech recognition apparatus 100 includes a speech recognition database (DB) 110, a marker insertion unit 120, a speech input unit 130, a speech recognition processing unit 140, a display unit 150, an operation input unit 160, a confirmation processing unit 170, and a confirmation. A result use unit 180 is included.

  The speech recognition database 110 stores in advance an acoustic model 111, a language model 112, and a dictionary 113, which are information used for speech recognition processing. The acoustic model 111 is obtained by converting a stochastic association between a voice feature quantity and a phonetic symbol into data. The dictionary 113 describes a plurality of text arrays as a candidate group of speech recognition results by speech recognition processing. The language model 112 is obtained by converting the appearance probability and the connection probability into data for each text array described in the dictionary 113.

  In the present embodiment, the text array described by the dictionary 113 is a syllable text array (hereinafter referred to as “word”) in which katakana words are written in katakana.

  Among a plurality of words (hereinafter referred to as “registered word group”) described by the dictionary 113, a word (exclusive text array, hereinafter referred to as “exclusive word”) having an exclusive section that is unique among the registered word groups ahead. )) Exists. In addition, among the registered words, a word that does not have an exclusive section (hereinafter referred to as a “non-exclusive word”) and that matches with one of the exclusive words in a predetermined number or more of syllables (hereinafter referred to as “forward match”). "Word"). In the present embodiment, the predetermined number is 3.

  The marker insertion unit 120 inserts a marker at an end position of an exclusive section of an exclusive word among words described in the dictionary 113. In addition, the marker insertion unit 120 inserts a marker at the end position of a non-exclusive word among the words described in the dictionary 113. This marker insertion process is performed before the voice recognition process is started.

  That is, the dictionary 113 describes a word with no marker inserted in the initial state. Thereafter, the dictionary 113 is in a state in which the exclusive word in which the marker is inserted is described by the function of the marker insertion unit 120. The inserted marker indicates the end position of the exclusive section of the exclusive word or the end position of the non-exclusive word.

  The voice input unit 130 inputs ambient voice, converts it into a voice signal, and outputs the voice signal to the voice recognition processing unit 140. When the user's uttered voice is included in the surrounding voice, the voice signal includes the user's uttered voice signal. The voice input unit 130 is, for example, a microphone provided for the mobile phone.

  The speech recognition processing unit 140 repeatedly performs speech recognition processing on the speech signal input from the speech input unit 130 at a predetermined cycle. In this speech recognition process, speech recognition is performed on a speech data portion (speech portion) that has already been input, and a text array composed of one or a plurality of syllables or a registered word in the dictionary 113 is converted into speech data (speech data). ) Is used as a provisional speech recognition result. The voice recognition processing unit 140 includes an acoustic analysis unit 141 and a recognition decoder unit 142.

  The acoustic analysis unit 141 analyzes the audio signal, and extracts a feature amount of speech (hereinafter referred to as “speech feature amount”) included in the speech data. Specifically, the acoustic analysis unit 141 performs frame processing on the audio signal, performs predetermined processing including Fourier analysis for each frame, and extracts cepstrum parameters and the like. And the acoustic analysis part 141 detects the audio | voice area from which the speech sound is contained from an analysis result, and produces | generates the time series data only by the audio | voice feature-value of an audio | voice area.

  The recognition decoder unit 142 refers to the acoustic model 111, the dictionary 113, and the language model 112 of the speech recognition database 110 based on the time series data of the speech feature amount generated by the acoustic analysis unit 141, and obtains a provisional speech recognition result. decide.

  More specifically, the recognition decoder unit 142 first converts the time-series data of speech feature values into a text array composed of one or more syllables. If the obtained text array does not exist in the dictionary 113, the text array is determined as a provisional speech recognition result. On the other hand, when the obtained text array exists in the dictionary 113, the corresponding word (word with the marker inserted) is determined as a provisional speech recognition result. Then, the recognition decoder unit 142 outputs the determined provisional speech recognition result (hereinafter referred to as “provisional result”) to the display unit 150 and the confirmation processing unit 170.

  That is, from the speech recognition processing unit 140, for example, while the user is speaking to the above-described microphone, a provisional result of speech text input is output at a predetermined cycle. For example, if the word that the user is trying to input speech text (hereinafter “desired word”) is a relatively long word, several provisional results can be obtained before the utterance of the word is completed become. In addition, as a provisional result, any of the following three types is possible: a text array that is not a registered word, an exclusive word in which a marker is inserted at the end position of the exclusive section, and a non-exclusive word in which a marker is inserted at the end position of the word. Will be obtained.

  Note that the speech recognition processing unit 140 calculates a certainty factor indicating the certainty of the provisional result for each provisional result. Then, the voice recognition processing unit 140 adds the calculated certainty factor to the provisional result output to the confirmation processing unit 170. The certainty factor is calculated based on, for example, the probability of the phonetic symbol with respect to the voice feature amount, the probability of appearance of the word with respect to the phonetic symbol, and the connection probability, and the probability of the best path (first candidate) and other paths (second This is performed using information such as probability difference after the candidate).

  The display unit 150 displays the provisional result input from the voice recognition processing unit 140. In addition, every time a new provisional result is input, the display unit 150 updates the display content with the new provisional result. The display unit 150 is, for example, a liquid crystal display provided for the mobile phone.

  The operation input unit 160 receives a determination operation for the provisional result displayed on the display unit 150. When the determination operation is performed, the operation input unit 160 notifies the confirmation processing unit 170 to that effect. The operation input unit 160 is, for example, a touch panel provided for the mobile phone.

  That is, when the latest provisional result is a desired word, the user can perform a determination operation on the provisional result.

  When the determination operation is performed by the operation input unit 160, the confirmation processing unit 170 uses the provisional result of the determination operation as a final voice recognition result (hereinafter referred to as “confirmation”) for the voice input by the voice input unit 130. Results ”).

  In addition, when the provisional result is an exclusive word, the confirmation processing unit 170 sets the exclusive word as the confirmation result even if the determination operation is not performed in the operation input unit 160. That is, when the provisional result is a text array having an exclusive section, the confirmation processing unit 170 determines the provisional result as the confirmation result before the determination operation is performed.

  However, in the present embodiment, the confirmation processing unit 170 sets the exclusive word as a confirmed result on the condition that all of the predetermined number of other provisional results obtained continuously in succession are forwardly matched. In the present embodiment, the predetermined number is 2. In other words, when a provisional result that is coincident with the front three times is obtained, the confirmation processing unit 170 sets the third provisional result as a confirmation result. This takes into account that misrecognition may occur due to the ambiguity of speech.

  Furthermore, even when the provisional result is a non-exclusive word, the confirmation processing unit 170 determines that the third provisional result is obtained when it coincides with all of the predetermined number of other provisional results obtained in succession. Is the final result. This takes into account when you have finished speaking a word.

  That is, when the registered word is obtained as a probable provisional result, the confirmation processing unit 170 determines the provisional result as the confirmation result without waiting for the determination operation. Then, the confirmation processing unit 170 outputs the determined confirmation result to the confirmation result use unit 180.

  The determination processing unit 170 excludes words other than the forward matching words, that is, words that do not match the exclusive words and words that do not match the exclusive words only in two or less syllables, from the determination of the above condition. And

  Whether or not the provisional result is an exclusive word is determined, for example, based on the presence or absence of the marker. Further, whether or not the provisional result is a forward matching word is determined, for example, based on the presence or absence of the marker.

  The confirmation result using unit 180 performs a predetermined process using the input confirmation result. The confirmation result using unit 180 is, for example, an e-mail application mounted on the mobile phone. In this case, the predetermined process is, for example, a process of displaying the input content on the display unit 150 while inputting the input confirmation result to an e-mail destination.

  Although not shown, the speech recognition apparatus 100 is, for example, a CPU (central processing unit), a storage medium such as a ROM (read only memory) storing a control program, and a working memory such as a RAM (random access memory). Have In this case, the function of each unit described above is realized by the CPU executing the control program.

  The speech recognition apparatus 100 having such a configuration can use the provisional result as a final result without waiting for a determination operation when a probable provisional result is obtained.

  A relatively long word often has an exclusive section, and furthermore, the distance from the end of the exclusive section to the end of the word (time required for speech, time required for speech recognition processing) is often long. In addition, when the desired word is an exclusive word, the desired word is often determined as a provisional result at the time when the speech recognition processing for the speech portion in the exclusive section is completed.

  Accordingly, when the desired word is a relatively long word, the speech recognition apparatus 100 can obtain a correct determination result before the speech recognition processing for the speech of the entire word is completed. That is, the speech recognition apparatus 100 can realize high-accuracy and high-speed speech text input even for relatively long words.

  Furthermore, the speech recognition apparatus 100 having the above configuration does not make the exclusive word a final result when all of the two other provisional results obtained in succession are not forward matching words for the exclusive word. Can be.

  If the confirmation result is determined before the speech recognition process for the whole word speech is completed, an erroneous confirmation result may be obtained depending on the utterance state and the input speech state. On the other hand, if all of the two other provisional results obtained in succession most recently are forward matching words for the latest provisional result, the possibility that the latest provisional result is correct is very high.

  Therefore, the speech recognition apparatus 100 can reduce the possibility of erroneous recognition. That is, the speech recognition apparatus 100 can realize high-speed speech text input with the recognition accuracy further improved.

  In addition, when the non-exclusive word is obtained as a provisional result, the speech recognition apparatus 100 also determines the provisional result when it coincides with all of the other two other provisional results obtained in succession. Can be the result. As a result, the speech recognition apparatus 100 confirms the provisional result without waiting for the determination operation when a predetermined time (the time required for obtaining the same provisional result three times) has elapsed after the word is finished. Can be the result.

  Above, description about the structure of the speech recognition apparatus 100 is complete | finished.

  Next, the operation of the speech recognition apparatus 100 will be described. The speech recognition apparatus 100 first performs a marker insertion operation as an initial setting operation, and then performs a speech recognition operation.

  The marker insertion operation is an operation for inserting a marker into a registered word group stored in the dictionary 113. The speech recognition operation is an operation for realizing high-accuracy and high-speed speech text input using the dictionary 113 in which the marker is inserted.

  First, the marker insertion operation will be described.

  FIG. 2 is a flowchart illustrating an example of the marker insertion operation of the speech recognition apparatus 100.

  First, in step S1010, the marker insertion unit 120 divides each word of the registered word group by a syllable. And the marker insertion part 120 sorts each word divided | segmented by the syllable in the order of 50 syllables, and stores it in the syllable matrix which is a virtual storage area.

  FIG. 3 is a diagram illustrating an example of a registered word group before the marker insertion operation.

  As shown in FIG. 3, the registered word group 210 before the marker insertion operation is a set of a plurality of words. Each word is a katakana syllable text array.

  FIG. 4 is a diagram illustrating an example of a syllable matrix in which registered word groups are stored.

  As shown in FIG. 4, the syllable matrix 220 is a list in which registered word groups are arranged in the order of 50 syllables with the positions of syllables aligned. For example, in the syllable matrix, the column containing the syllable “ta” in the row where the word “Igza Tablet” is stored, and the syllable “phone” in the row where the word “Igzaphone” is stored next are stored. The assigned columns belong to the same column.

  In step S1020 of FIG. 2, the marker insertion unit 120 selects one row (word) from the syllable matrix. Here, it is assumed that the marker insertion unit 120 sequentially selects from the top to the bottom of the syllable matrix. For example, the marker insertion unit 120 selects a row in which the word “advance” is stored.

  In step S1030, the marker insertion unit 120 compares the selected row for each column (syllable) with the next row, and determines whether or not there is a forward match in at least one or more columns. The range corresponding to the front coincides with the maximum range among the ranges that coincide between the two sequences with reference to the front of each sequence.

  If the marker insertion unit 120 has a forward match between the selected row and the next row (S1030: YES), the marker insertion unit 120 proceeds to step S1050 described later. In addition, when the marker insertion unit 120 does not match the selected line with the next line (S1030: NO), the process proceeds to step S1040.

  For example, when the word “Ixaphone Triple X” is selected, there is no forward match with the word “Uni” on the next line. Therefore, in such a case, the process proceeds to step S1040.

  In step S1040, the marker insertion unit 120 proceeds to step S1050, which will be described later, when the front comparison also matches in the previous comparison (S1040: YES). If the marker insertion unit 120 is not forward in the previous comparison (S1040: NO), the marker insertion unit 120 proceeds to step S1060 described later.

  The case where there is a forward match in the previous comparison means that the word in the selected row is a forward match with both words in the preceding and succeeding rows. For example, when a line storing the word “Igza Tablet” is selected, the word “Igza” in the previous line and the word “Igzaphone” in the next line are “I” and “Gu”. , “The” is stored in the three columns in which “the” is stored, and the front coincides. Therefore, in this case, the process proceeds to step S1050.

  In step S1050, the marker insertion unit 120 adds, to the matched character string list, a part that matches the word in the next line of the word stored in the selected line. The matched character string list is a list of portions (that is, common portions) that are forward-matched with the next word in a character string that constitutes a forward-matched word between the two words before and after.

  FIG. 5 is a diagram illustrating an example of the matching character string list.

  As shown in FIG. 5, the matched character string list 230 describes only a portion that matches the next word in the character string that forms a forward matched word between the two words before and after. Therefore, for example, in the registered word group 210 shown in FIG. 3, the character string corresponding to “advance” is not described in the matching character string list 230.

  In step S1060, the marker insertion unit 120 determines whether the selected row is the last row (word) of the syllable matrix.

  If the selected row is not the last row of the syllable matrix (S1060: NO), the marker insertion unit 120 returns to step S1020 and moves to the process for the next row. When the marker insertion unit 120 completes the process for all the rows of the syllable matrix (S1060: YES), the marker insertion unit 120 proceeds to step S1070.

  In step S1070 in FIG. 2, the marker insertion unit 120 again selects one row (word) from the syllable matrix. Here, it is assumed that the marker insertion unit 120 sequentially selects from the top to the bottom of the syllable matrix. For example, the marker insertion unit 120 selects a row in which the word “advance” is stored.

  In step S1080, the marker insertion unit 120 determines whether or not the forward portion of the word in the selected line is included in the matched character string list (see FIG. 5). In other words, this determination is a determination as to whether or not the word is one of an exclusive word and a front matching character string.

  When the word on the selected line is included in the matched character string list (S1080: YES), the marker insertion unit 120 proceeds to step S1090 described later. In addition, when the word in the selected line is not included in the matched character string list (S1080: NO), the marker insertion unit 120 proceeds to step S1110 described later.

  For example, when the word “advance” is selected, the front part of this word is not included in the matching character string list 230 shown in FIG. Therefore, in this case, the process proceeds to step S1110. Further, for example, when the word “Igusa” is selected, the front part of this word is included in the matching character string list 230 shown in FIG. Therefore, in this case, the process proceeds to step S1090.

  In step S1090, the marker insertion unit 120 compares the selected row with the next row for each column (syllable), and determines whether or not there is a forward match in a predetermined number (3) or more columns.

  The marker insertion unit 120 proceeds to step S1100 when the number of columns equal to or greater than the predetermined number (3) is a forward match (S1090: YES). In addition, when the marker insertion unit 120 does not match the front in the predetermined number (3) or more (S1090: NO), the marker insertion unit 120 proceeds to step S1110 described later.

  For example, assume a case where a row in which the word “exa tablet” is stored is selected. This line corresponds to the line in which the word “exaphone” in the next line is stored in the range of “exa”. In other words, at least “Iguza” is included in “Iguzaphone”. Therefore, in this case, the process proceeds to step S1100.

  Further, for example, a case is assumed where a row in which the word “squid” is stored is selected. This line coincides with the line in which the word “Igza” in the next line is stored in the range of “i”, but the number of phonemes is less than the predetermined number (3). Therefore, the process proceeds to step S1110.

  In step S1100, the marker insertion unit 120 inserts a marker into the registered word group described in the dictionary 113 corresponding to the selected word, and the process proceeds to step S1120. More specifically, the marker insertion unit 120 inserts a marker at a position immediately before the foremost column among the columns (syllables) in which the selected row does not match the next row.

  On the other hand, in step S1110, the marker insertion unit 120 inserts a marker into the registered word group described in the dictionary 113 corresponding to the currently selected word, and proceeds to step S1120. More specifically, the marker insertion unit 120 inserts a marker at the end position of the word.

  In step S1120, the marker insertion unit 120 determines whether the currently selected row is the last row (word) of the syllable matrix.

  If the selected row is not the last row of the syllable matrix (S1120: NO), the marker insertion unit 120 returns to step S1070 and moves to the process for the next row. And the speech recognition apparatus 100 will complete | finish a marker insertion operation, if a process is completed about all the lines of a syllable matrix (S1120: YES).

  By such marker insertion operation, the speech recognition apparatus 100 changes the exclusive word from the registered word group described in the dictionary 113 to a state in which the marker is inserted at the end position of the exclusive section, and sets the non-exclusive word. , And a marker inserted at the end of the word. That is, all the registered words are in a state where a marker is inserted.

  Note that the speech recognition apparatus 100 uses, for example, a character string “@” + reading other additional information + “@” as a marker. This additional information is used by, for example, the determination result use unit 180.

  FIG. 6 is a diagram illustrating an example of a registered word group after the marker insertion operation.

  As shown in FIG. 6, the registered word group 240 after the marker insertion operation is in a state where a marker is inserted. For example, a marker “@ ixablelet @” is inserted between “ta” and “bu” in the word “exa tablet”.

  This is the end of the description of the marker insertion operation.

  Next, the voice recognition operation will be described.

  FIG. 7 is a flowchart showing an example of the voice recognition operation. Here, it is assumed that the voice input unit 130 continuously performs voice input and outputs voice data to the voice recognition processing unit 140. In addition, the speech recognition apparatus 100 performs a speech recognition operation described below every time the above-described speech section is started and every time a determination result is obtained during the speech section.

  First, in step S2010, the speech recognition processing unit 140 performs the above-described speech recognition on a portion of speech data that has already been input since the start of the current speech recognition operation (hereinafter referred to as “already input data portion”). Process. That is, the speech recognition processing unit 140 determines a provisional result of speech recognition, and calculates the certainty factor of the provisional result in the process. Then, the voice recognition processing unit 140 outputs the determined provisional result to the display unit 150 and causes the display unit 150 to display it.

  It is assumed that the word is displayed on display unit 150 with the marker removed. The marker 150 is removed by the display unit 150, for example. In addition, the voice recognition processing unit 140 outputs the determined provisional result to the confirmation processing unit 170 together with the certainty factor.

  The speech recognition processing unit 140 performs the process of step S2010 at a predetermined cycle (for example, 100 msec) by returning the process from step S2100 described later. Therefore, the provisional result is obtained repeatedly while the utterance is performed. In addition, the already-input data portion for which the final result has not been obtained becomes longer as time elapses. Thus, the provisional results obtained can change over time.

  In step S2020, the confirmation processing unit 170 determines whether the input certainty factor is equal to or greater than a first predetermined value. This predetermined value is a value determined by experiments or the like based on the accuracy and speed required for speech text input, for example.

  If the certainty factor is greater than or equal to the predetermined value (S2020: YES), the confirmation processing unit 170 proceeds to step S2030. Further, when the certainty factor is less than the predetermined value (S2020: NO), the confirmation processing unit 170 proceeds to step S2050 described later.

  In step S2030, the confirmation processing unit 170 determines whether the provisional result includes a marker. That is, the confirmation processing unit 170 determines whether or not the provisional result is a registered word.

  If the provisional result includes a marker (S2030: YES), the confirmation processing unit 170 proceeds to step S2040 described later. If the provisional result does not include a marker (S2030: NO), the confirmation processing unit 170 proceeds to step S2050.

  For example, when the quality of the input voice is low or the provisional result is “I”, the process proceeds to step S2050 described later. Further, for example, when the quality of the input voice is high, or when the provisional result is “Ixa @ ixa @” or “Ixata @ ixatable @ Brett”, the process proceeds to Step S2040.

  In step S2040, the confirmation processing unit 170 determines whether or not the current temporary result matches the previous recording. Here, the previous record is the latest one of the past provisional results that have been recorded in the past in step S2090 and maintained. Note that the confirmation processing unit 170 may proceed to step S2090 without executing steps S2020 to S2080 for the initial provisional result of the speech section.

  The confirmation processing unit 170 proceeds to step S2060 when the provisional result of this time coincides with the previous recording (S2040: YES). If the current provisional result does not coincide with the previous record (S2040: NO), the confirmation processing unit 170 proceeds to step S2070 described later.

  For example, if the previous record is “Igusa @ ixa @” and the current provisional result is “Exact @ tablet @ Brett”, the process proceeds to step S2060.

  In step S2060, the confirmation processing unit 170 increments the front matching parameter. The forward coincidence parameter is a parameter indicating the number of times the provisional result that is forward coincidence is obtained continuously with a predetermined number (3) or more (hereinafter referred to as “continuous forward coincidence number”), and the initial value is 0 .

  For example, a case is assumed in which the provisional result of the previous round is “squid @ bandit @”, the previous provisional result is “igza @ ixa @”, and the current provisional result is “exact @ tablet @ brett”. In this case, the forward match parameter is “1”.

  In step S2080, confirmation processing section 170 determines whether or not the forward matching parameter has reached a predetermined value (2). In other words, the confirmation processing unit 170 satisfies the condition that the latest provisional result is forwardly coincident with all of two or more other provisional results obtained consecutively most recently (hereinafter referred to as “count condition”). To determine whether or not

  If the forward matching parameter reaches the predetermined value (2) (S2080: YES), the confirmation processing unit 170 proceeds to step S2110 described later. If the forward matching parameter has not reached the predetermined value (2) (S2080: NO), the confirmation processing unit 170 proceeds to step S2090.

  In step S2090, the confirmation processing unit 170 records the current provisional result.

  On the other hand, in step S2050, if there is a previous record, the confirmation processing unit 170 clears this and proceeds to step S2100 described later. That is, when the certainty factor of the provisional result is low or the provisional result is not a registered word, the confirmation processing unit 170 excludes the previous and current provisional results from the above-described count condition determination.

  In step S2070, the confirmation processing unit 170 initializes the forward matching parameter and proceeds to step S2100. That is, if the current temporary result does not coincide with the previous record, the confirmation processing unit 170 shifts the starting point for counting the number of consecutive forward matches to the next temporary result.

  In step S2100, the confirmation processing unit 170 determines whether a determination operation has been performed on the current provisional result. That is, the confirmation processing unit 170 determines whether or not the operation input unit 160 has notified that the determination operation has been performed.

  If the determination operation is not performed (S2100: NO), the confirmation processing unit 170 returns to step S2010, and proceeds to processing for the already input data portion including the newly input data portion. In addition, when the determination operation is performed (S2100: YES), the confirmation processing unit 170 proceeds to step S2110.

  For example, when the number of consecutive forward matches has reached 2, even if the determination operation is not performed, the process proceeds to step S2110. This indicates that the possibility that the exclusive section of the exclusive word has been recognized or that the recognition of the entire word has been completed is sufficiently high.

  If a determination operation is performed, the process proceeds to step S2110 even if the number of consecutive forward matches has not reached 2.

  In step S <b> 2110, the confirmation processing unit 170 determines the current provisional result as a confirmation result, and outputs the confirmation result to the confirmation result use unit 180. Then, the voice recognition device 100 ends the voice recognition operation.

  The confirmation result using unit 180 may use a confirmation result in a state where the marker is removed, or may use a confirmation result in a state where the marker is included. The removal of the marker may be performed by the confirmation processing unit 170 or the confirmation result using unit 180.

  With such a speech recognition operation, the speech recognition apparatus 100 can periodically obtain a provisional result during utterance using the dictionary 113 in which a word with a marker inserted is registered. In addition, the speech recognition apparatus 100 can determine whether the utterance of the exclusive section of the exclusive word or the utterance of the entire word is sufficiently high even if the determination operation is not performed. The provisional word obtained at the time can be used as the final result.

  This is the end of the description of the voice recognition operation.

  Next, a specific example of the utterance content and the provisional result will be described, and the state of operation until the final result is obtained and the effect of the present invention will be described.

  FIG. 8 is a diagram for explaining an example of the time required from the start of utterance to the determination result being obtained when a relatively long word voice is input.

  As shown in the upper side of FIG. 8, it is assumed that the word “exaphone triple X” is uttered. Then, it is assumed that the utterance start time is time t0 and utterance end time t2. That is, it is assumed that the word “exaphone triple X” requires time t2 to speak. Also, it is assumed that time t1 is required from the start of voice input until the provisional result of the corresponding portion is obtained, due to processing such as buffering of voice data.

  When the final result is obtained based on the entire voice data of the speech voice of the word “Ixaphone Triple X”, the accuracy of the final result is high. However, in this case, the final result is obtained at time t1 + t2 when time t1 + t2 has elapsed from time t0. That is, conventionally, the speaker waits for the provisional result to be displayed until after time t1 + t2, and then performs the confirmation operation.

  Here, it is assumed that “exaphone triple X” is an exclusive word “exa font @ixaphonetriple X @ripple X” as shown in FIG. In this case, as shown in the lower side of FIG. 8, the speech recognition apparatus 100 sets the provisional result as the final result at the stage of time te when the provisional result for the already input data portion up to the position “g” is obtained. be able to. The exclusive section “exa font” does not match any other word. Therefore, the speech recognition apparatus 100 can determine the speech recognition result with relatively high accuracy at time te, which is a time earlier than time t1 + t2.

  However, the speech recognition accuracy when only a part of the speech data is used is lower than the speech recognition accuracy when the entire speech data is used. Therefore, as described above, the speech recognition apparatus 100 determines that the provisional result is an exclusive word, and that the provisional result is a forward match with all of the predetermined number of other provisional results obtained in succession. This is a condition for finalization.

  FIG. 9 and FIG. 10 are diagrams illustrating an example of a provisional result obtained at each time and an example of how the result is determined. FIG. 9 is an example when the quality of the input voice is good, and FIG. 10 is an example when the quality of the input voice is not good. In any case, it is assumed that the desired word is “Ixaphone Triple X”.

  As shown in FIG. 9, at time t11 to t14, provisional results of “squid @ bandit @”, “exa @ ixa @”, “exaphone @ ixaphone @”, and “exafont @ ixaphonetripleX @ ripple X” are obtained, respectively. Shall be. When the quality of the input speech is good, a word that matches the desired word in front is often a provisional result.

  In “Igusa @ ixa @”, “Ixaphone @ ixaphone @”, and “Ixafont @ ixaphonetripleX @ RippleX”, as indicated by the underline in FIG. In addition, these provisional results are continuous, and “exafont @ ixaphonetripleX @ ripple X” is an exclusive word.

  Therefore, the voice recognition result is determined when “exafont @ ixaphoneetriple @ ripplex” is obtained at time t14. This is a time earlier than the time te when the final result is obtained based on the entire voice data of the speech voice of the word “Ixaphone Triple X”.

  Also, as shown in FIG. 10, at time t11 to t16, “squid @ bandit @”, “igza @ ixa @”, “i”, “squid @ bandit @”, “exaphones @ ixaphonesports @ ports”, respectively. Suppose that a provisional result “exafont @ ixaphonetrixX @ ripple X” is obtained. Further, it is assumed that the certainty of the provisional result “squid @ bandit @” at time t14 is low. When the quality of the input speech is not good, a word that does not coincide with the desired word is often a provisional result or the certainty level is low.

  “Igza @ ixa @” is a forward matching word of “Ixafont @ ixaphonetripleX @ Ripple X”, but “I” is not a forward matching word of “Igzafont @ ixaphonetripleX @ RippleX”. However, the record of “I” is cleared because the certainty of the next “squid @ bandit @” is low. Also, “squid @ bandit @” is not recorded. Therefore, these are not subject to the above conditions.

  Therefore, after “Ixaphones @ ixaphonesports @ Ports”, the voice recognition result is determined when “Ixafont @ ixaphonetripleX @ RippleX” is obtained as a provisional result at time t16. This is a time earlier than the time te when the final result is obtained based on the entire voice data of the speech voice of the word “Ixaphone Triple X”.

  That is, the speech recognition apparatus 100 can confirm the speech recognition result when the speech recognition result is narrowed down to one before the speaker finishes speaking all the desired words. That is, the speech recognition apparatus 100 can realize high-speed response speech recognition.

  However, when the quality of the input voice is poor, the provisional result fluctuates greatly, and the above condition may not be satisfied even at time te. In such a case, for example, the speech recognition result is determined by the user's determination operation or correction operation.

  As described above, when the fluctuation of the provisional result is small, the speech recognition apparatus 100 performs early determination of the result, and when the fluctuation of the provisional result is large, the reliability of the provisional recognition result is sufficiently high. Then, confirm the result. Therefore, the speech recognition apparatus 100 can achieve high speed and high accuracy of speech text input with a good balance, especially for relatively long words.

  This is the end of the description of the operations and effects of the specific example.

  As described above, in the speech recognition apparatus 100 according to the present embodiment, the provisional result is an exclusive word, and the predetermined number of other provisional results obtained in succession most recently coincide with each other. As a condition, the exclusive word is taken as the final result. Thereby, the speech recognition apparatus 100 can realize high-accuracy and high-speed speech text input even for relatively long words.

  In the embodiment described above, the registered word is determined using the marker, but the determination method of these words is not limited to this. For example, the speech recognition apparatus may determine a registered word using word grouping or other additional information.

  In addition, each functional unit illustrated in FIG. 1 is not necessarily provided integrally in one apparatus. For example, the voice recognition database and the marker insertion unit may be arranged on a server on the Internet, and the other functional units may be arranged on a mobile phone.

  Further, the threshold value (predetermined value) for determining the forward match with the exclusive word and the threshold value (predetermined value) for determining the number of consecutive forward matches are not limited to the above example. These values are determined by experiments or the like based on the accuracy and speed required for speech text input, for example.

  Further, the format and contents of the text array described in the dictionary as candidates for the speech recognition result are not limited to the above example. For example, the text array may be a kana / kanji character string. The text array may be a word and a sentence (phrase).

  The apparatus to which the present invention is applied is not limited to the above-described example. The present invention can be applied to various electronic devices such as a tablet terminal and a personal computer.

  INDUSTRIAL APPLICABILITY The present invention is useful as a speech recognition apparatus and a speech recognition result determination method that can realize high-accuracy and high-speed speech text input even for relatively long words.

DESCRIPTION OF SYMBOLS 100 Speech recognition apparatus 110 Speech recognition database 111 Acoustic model 112 Language model 113 Dictionary 120 Marker insertion part 130 Speech input part 140 Speech recognition process part 141 Acoustic analysis part 142 Recognition decoder part 150 Display part 160 Operation input part 170 Confirmation process part 180 Confirmation Results use department

Claims (10)

A speech recognition database that stores a dictionary describing multiple text arrays;
A voice input unit for inputting voice;
A voice recognition processor to I line speech recognition, the result provisional speech recognition results of speech recognition on the audio portion which has already been input among the speech,
When the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead, the confirmation processing unit that uses the exclusive text array as a confirmed speech recognition result for the speech and, the possess,
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The confirmation processing unit confirms that the provisional speech recognition result is the exclusive text array and is forward-matched with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. As a condition, the exclusive text array is the final speech recognition result,
The predetermined number is two;
Voice recognition device. A speech recognition database that stores a dictionary describing multiple text arrays;
A voice input unit for inputting voice;
A voice recognition processing unit that performs voice recognition on a voice part that has already been input in the voice and sets the voice recognition result as a provisional voice recognition result;
When the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead, the confirmation processing unit that uses the exclusive text array as a confirmed speech recognition result for the speech And having
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The confirmation processing unit confirms that the provisional speech recognition result is the exclusive text array and is forward-matched with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. As a condition, the exclusive text array is the final speech recognition result,
The text array is a syllable text array;
The confirmation processing unit further excludes the text arrangement that does not coincide with the exclusive text arrangement in the predetermined number of syllables or more from the judgment of the condition.
Voice recognition device. A speech recognition database that stores a dictionary describing multiple text arrays;
A voice input unit for inputting voice;
A voice recognition processing unit that performs voice recognition on a voice part that has already been input in the voice and sets the voice recognition result as a provisional voice recognition result;
When the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead, the confirmation processing unit that uses the exclusive text array as a confirmed speech recognition result for the speech And having
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The confirmation processing unit confirms that the provisional speech recognition result is the exclusive text array and is forward-matched with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. As a condition, the exclusive text array is the final speech recognition result,
The exclusive text array is a syllable text array in which a marker indicating the end position of the exclusive section is inserted,
Of the plurality of text arrays, the text array other than the exclusive text array is a syllable text array in which a marker indicating the end position of the text array is inserted,
When the provisional speech recognition result is one in which the marker is not inserted, the confirmation processing unit excludes the provisional speech recognition result from the determination of the condition.
Voice recognition device. A speech recognition database that stores a dictionary describing multiple text arrays;
A voice input unit for inputting voice;
A voice recognition processing unit that performs voice recognition on a voice part that has already been input in the voice and sets the voice recognition result as a provisional voice recognition result;
When the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead, the confirmation processing unit that uses the exclusive text array as a confirmed speech recognition result for the speech A speech recognition device having:
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The confirmation processing unit confirms that the provisional speech recognition result is the exclusive text array and is forward-matched with all of the predetermined number of other provisional speech recognition results obtained consecutively most recently. As a condition, the exclusive text array is the final speech recognition result,
The speech recognition apparatus sorts a plurality of text arrays described in the dictionary in the order of front matching, and performs a process of determining a front matching range for each of the adjacent text array pairs. A marker insertion unit for inserting a marker into the exclusive text array.
Voice recognition device. A display unit for displaying the provisional voice recognition result;
An operation input unit that receives a determination operation on the provisional voice recognition result displayed on the display unit;
The confirmation processing unit
When the determination operation is performed, the provisional speech recognition result on which the determination operation is performed is the final speech recognition result, and when the provisional speech recognition result is the exclusive text array, the determination operation is Even if it is not performed, the exclusive text array is set as the confirmed speech recognition result.
Speech recognition apparatus according to claim 1. The voice recognition processing unit
For each provisional speech recognition result, calculate a certainty factor indicating the certainty of the provisional speech recognition result,
The confirmation processing unit
The provisional speech recognition result with the certainty factor less than a predetermined value is excluded from the determination of the condition.
Speech recognition apparatus according to claim 1. A speech recognition database storing a dictionary describing a plurality of text arrays; a speech input unit for inputting speech; and speech recognition for speech portions already input in the speech, and results of speech recognition A speech recognition result determination method in a speech recognition device having a provisional speech recognition result,
Determining whether the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead;
When the provisional speech recognition result is the exclusive text array, the exclusive text array is set as a confirmed speech recognition result for the speech, and
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The step of setting the final speech recognition result includes the provisional speech recognition result being the exclusive text array, and a predetermined number of the other provisional speech recognition results obtained in succession and the front. On the condition that they match, the exclusive text array is set as the confirmed speech recognition result,
The predetermined number is two;
Voice recognition result confirmation method . A speech recognition database storing a dictionary describing a plurality of text arrays; a speech input unit for inputting speech; and speech recognition for speech portions already input in the speech, and results of speech recognition A speech recognition result determination method in a speech recognition device having a provisional speech recognition result,
Determining whether the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead;
When the provisional speech recognition result is the exclusive text array, the exclusive text array is set as a confirmed speech recognition result for the speech, and
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The step of setting the final speech recognition result includes the provisional speech recognition result being the exclusive text array, and a predetermined number of the other provisional speech recognition results obtained in succession and the front. On the condition that they match, the exclusive text array is set as the confirmed speech recognition result,
The text array is a syllable text array;
In the step of determining the voice recognition result, the text array that does not coincide with the exclusive text array in the predetermined number or more of syllables is excluded from the determination of the condition.
Voice recognition result confirmation method. A speech recognition database storing a dictionary describing a plurality of text arrays; a speech input unit for inputting speech; and speech recognition for speech portions already input in the speech, and results of speech recognition A speech recognition result determination method in a speech recognition device having a provisional speech recognition result,
Determining whether the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead;
When the provisional speech recognition result is the exclusive text array, the exclusive text array is set as a confirmed speech recognition result for the speech, and
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The step of setting the final speech recognition result includes the provisional speech recognition result being the exclusive text array, and a predetermined number of the other provisional speech recognition results obtained in succession and the front. On the condition that they match, the exclusive text array is set as the confirmed speech recognition result,
The exclusive text array is a syllable text array in which a marker indicating the end position of the exclusive section is inserted,
Of the plurality of text arrays, the text array other than the exclusive text array is a syllable text array in which a marker indicating the end position of the text array is inserted,
The step of setting the final voice recognition result is such that when the provisional voice recognition result is one in which the marker is not inserted, the provisional voice recognition result is excluded from the determination of the condition.
Voice recognition result confirmation method. A speech recognition database storing a dictionary describing a plurality of text arrays; a speech input unit for inputting speech; and speech recognition for speech portions already input in the speech, and results of speech recognition A speech recognition result determination method in a speech recognition device having a provisional speech recognition result,
Determining whether the provisional speech recognition result is an exclusive text array having an exclusive section that is unique among the plurality of text arrays ahead;
When the provisional speech recognition result is the exclusive text array, the exclusive text array is set as a confirmed speech recognition result for the speech, and
The voice recognition processing unit repeatedly performs the voice recognition at a predetermined cycle,
The step of setting the final speech recognition result includes the provisional speech recognition result being the exclusive text array, and a predetermined number of the other provisional speech recognition results obtained in succession and the front. On the condition that they match, the exclusive text array is set as the confirmed speech recognition result,
The speech recognition result determination method performs a process of sorting a plurality of text arrays described in the dictionary in a front matching order and determining a front matching range for each of the adjacent text array pairs. Further comprising inserting a marker into the exclusive text array.
Voice recognition result confirmation method.

Images