Delimiter insertion device and speech recognition system

Abstract

This delimiter insertion device comprises: a word-spacing time acquisition unit that acquires a word-spacing time, which is the length of time until the next word is uttered, for each word included in uttered speech; and a delimiter insertion unit that, on the basis of a delimiter insertion model and the word-spacing time, inserts a delimiter into target text, which is text obtained through speech recognition of the uttered speech. The delimiter insertion model outputs delimiter inference information indicating a delimiter in response to input of a delimiter removal sentence. The delimiter insertion unit inserts a delimiter into the target text on the basis of the delimiter inference information obtained by inputting the target text into the delimiter insertion model.

Description

Delimiter insertion device and speech recognition system

[0001] The present invention relates to a delimiter insertion device and a speech recognition system.

[0002] Techniques for inserting punctuation marks into text obtained by speech recognition are known. For example, Patent Document 1 discloses a technique in which punctuation marks are inserted into text by an engine trained with training data in the form of statistically punctuated text.

[0003] JP 2012-508903 A

[0004] In a typical speech recognition engine, after acquiring text consisting of a string of words from a spoken voice, delimiters such as punctuation marks are inserted at appropriate positions in a sentence. When delimiters are inserted by referring only to text information, there are cases where the insertion positions of the delimiters are not incorrect as a sentence but are different from the speaker's intention.

[0005] Therefore, the present invention has been made in consideration of the above problems, and aims to insert delimiters into text obtained by speech recognition processing of spoken voice at positions intended by the speaker.

[0006] In order to solve the above problem, a punctuation mark insertion device according to one aspect of the present disclosure is a punctuation mark insertion device that inserts punctuation marks that separate sentences after words included in text obtained by speech recognition of spoken speech, and includes: a space-time acquisition unit that acquires an inter-word time, which is the length of time until the next word is spoken for each word included in the spoken speech; and a punctuation mark insertion unit that inserts punctuation marks into target text, which is text obtained by speech recognition of the spoken speech, based on a punctuation mark insertion model and the inter-word times. The punctuation mark insertion model receives as input at least a punctuation mark-removed sentence, which is a sentence that does not contain punctuation marks, and outputs punctuation mark prediction information indicating punctuation marks to be inserted after each word included in the punctuation mark-removed sentence. The punctuation mark insertion model is a model generated by machine learning using training data that includes pairs of the punctuation mark-removed sentence and a punctuation mark-containing sentence, which is a sentence that includes punctuation marks. The punctuation mark insertion unit inserts punctuation marks into the target text based on the punctuation mark prediction information that has been adjusted according to the punctuation mark prediction information, which is obtained by inputting the target text as a punctuation mark-removed sentence into the punctuation mark insertion model.

[0007] According to the above aspect, an interword time in a spoken voice is acquired. The interword time reflects the intention of the speaker when speaking. Then, by inputting the target text into a punctuation symbol insertion model, punctuation symbol prediction information is acquired that is obtained and adjusted according to the space information. The acquired punctuation symbol prediction information is information that has been adjusted according to the interword time of each word in the spoken voice, and therefore indicates punctuation symbols that reflect the intention of the speaker. Then, by inserting punctuation symbols into the target text based on the punctuation symbol prediction information, it is possible to obtain text in which punctuation symbols are inserted at positions that conform to the intention of the speaker.

[0008] It is possible to insert delimiters into the text obtained by speech recognition processing of the spoken voice at the positions intended by the speaker.

[0009] 1 is a block diagram showing the functional configuration of the delimiter insertion device of this embodiment. FIG. 2 is a hardware block diagram of the delimiter insertion device. FIG. 3 is a diagram explaining a problem solved by the delimiter insertion device of this embodiment. FIG. 4 is a diagram showing a process of acquiring a target text. FIG. 5 is a diagram showing a first example of training data used for machine learning of the delimiter insertion model. FIG. 6 is a diagram showing a first example of the configuration of the delimiter insertion model. FIG. 7 is a diagram showing an example of adjustment rule information referenced for adjusting delimiter prediction information based on interword time. FIG. 8 is a diagram showing an example of a process of adjusting delimiter prediction information. FIG. 9 is a diagram showing an example of a process of correcting interword time. FIG. 10 is a diagram showing a second example of training data used for machine learning of the delimiter insertion model. FIG. 11 is a diagram showing a second example of the configuration of the delimiter insertion model. FIG. 12 is a diagram showing a third example of training data used for machine learning of the delimiter insertion model. A functional block diagram showing an example of the configuration of a speech recognition system of this embodiment. A flowchart showing the processing contents of a delimiter insertion method in the delimiter insertion device. A diagram showing the configuration of a delimiter insertion program.

[0010] The present invention will be described with reference to the accompanying drawings, in which like reference numerals are used to denote like parts, and a description thereof will be omitted.

[0011] 1 is a diagram showing the functional configuration of a delimiter insertion device according to this embodiment. The delimiter insertion device according to this embodiment is a device that inserts delimiters, which separate sentences, after words included in text obtained by speech recognition of spoken speech.

[0012] In this embodiment, an example will be described in which the delimiter insertion device 10 inserts delimiters such as periods, commas, and question marks into English text, but the present invention is not limited to this example. The delimiter insertion device 10 may be a device that inserts delimiters such as periods and commas into Japanese text, or a device that inserts delimiters specific to another language into text in that language.

[0013] 1, the delimiter insertion device 10 functionally comprises a text acquisition unit 11, an inter-word time acquisition unit 12, a delimiter insertion unit 13, and an output unit 14. These functional units 11 to 14 may be configured in a single device, or may be distributed across multiple devices.

[0014] The block diagram shown in FIG. 1 shows functional blocks. These functional blocks (components) are realized by any combination of hardware and / or software. Furthermore, the method for realizing each functional block is not particularly limited. That is, each functional block may be realized using a single device that is physically or logically coupled, or may be realized using two or more physically or logically separated devices that are directly or indirectly connected (e.g., wired, wireless, etc.) and these multiple devices. The functional block may also be realized by combining software with the single device or multiple devices.

[0015] Functions include, but are not limited to, judgment, determination, assessment, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, and assignment. For example, a functional block (component) that performs transmission is called a transmitting unit or transmitter. As mentioned above, there are no particular limitations on how these functions are implemented.

[0016] For example, the delimiter insertion device 10 according to an embodiment of the present invention may function as a computer. Fig. 2 is a diagram showing an example of the hardware configuration of the delimiter insertion device 10 according to this embodiment. The delimiter insertion device 10 may be physically configured as a computer including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, etc.

[0017] In the following description, the term "apparatus" may be interpreted as a circuit, a device, a unit, etc. The hardware configuration of the delimiter insertion apparatus 10 may be configured to include one or more of the devices shown in the figure, or may be configured to exclude some of the devices.

[0018] Each function of the delimiter insertion device 10 is realized by loading specified software (programs) onto hardware such as the processor 1001 and memory 1002, causing the processor 1001 to perform calculations and control communication via the communication device 1004 and the reading and / or writing of data in the memory 1002 and storage 1003.

[0019] The processor 1001 controls the entire computer by running, for example, an operating system. The processor 1001 may be configured as a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. For example, the functional units 11 to 14 shown in FIG. 1 may be realized by the processor 1001.

[0020] The processor 1001 also reads programs (program codes), software modules, and data from the storage 1003 and / or the communication device 1004 into the memory 1002 and executes various processes in accordance with these. The programs used are those that cause a computer to execute at least some of the operations described in the above-described embodiments. For example, the functional units 11 to 15 of the delimiter insertion device 10 may be implemented by a control program stored in the memory 1002 and running on the processor 1001. While the above-described various processes have been described as being executed by one processor 1001, they may also be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented on one or more chips. The programs may also be transmitted from a network via a telecommunications line.

[0021] The memory 1002 is a computer-readable recording medium and may be composed of at least one of, for example, a read-only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), and a random access memory (RAM). The memory 1002 may also be called a register, a cache, a main memory (primary storage device), or the like. The memory 1002 can store executable programs (program codes), software modules, and the like for implementing the pseudo data generation method and sentence generation method according to one embodiment of the present invention.

[0022] Storage 1003 is a computer-readable recording medium, and may be, for example, at least one of an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, etc. Storage 1003 may also be referred to as an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, a server, or other appropriate medium including memory 1002 and / or storage 1003.

[0023] The communication device 1004 is hardware (transmission / reception device) for communicating between computers via a wired and / or wireless network, and is also called, for example, a network device, a network controller, a network card, or a communication module.

[0024] The input device 1005 is an input device (e.g., a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives input from the outside. The output device 1006 is an output device (e.g., a display, a speaker, an LED lamp, etc.) that outputs to the outside. Note that the input device 1005 and the output device 1006 may be integrated into one device (e.g., a touch panel).

[0025] Furthermore, each device such as the processor 1001 and the memory 1002 is connected to a bus 1007 for communicating information. The bus 1007 may be configured as a single bus, or may be configured as different buses between the devices.

[0026] The delimiter insertion device 10 may also be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented by at least one of these pieces of hardware.

[0027] The problem to be solved by the delimiter insertion device 10 of this embodiment will be described with reference to FIG. 3. The speech speeches sp11 and sp21 shown in FIG. 3 are each uttered by a speaker with a different intention. The speech recognition results sr11 and sr21 obtained by speech recognition of the speech speeches sp11 and sp21 are the same text, "I know it's been there forever." In the speech speech sp11, the inter-word time until the word following the word "know" is uttered is 1.2 seconds. On the other hand, in the speech speech sp21, the inter-word time until the word following the word "know" is uttered is 0.1 seconds.

[0028] In conventional delimiter insertion techniques, delimiters are inserted into text obtained as speech recognition results at positions that are appropriate for a sentence. Therefore, speech recognition results sr12 and sr22 obtained by inserting delimiters into speech recognition results sr11 and sr21 using conventional delimiter insertion techniques are identical, even though the original spoken speech is different.

[0029] The speech recognition result sr22 has a period after the word "forever", as intended by the speaker of the utterance sp21. In contrast, although the utterance sp11 was intended to have a period inserted after the word "know", the speech recognition result sr12 has a period after the word "forever". The position of this period is different from the position intended by the speaker.

[0030] The delimiter insertion device 10 of this embodiment inserts delimiters using inter-word times in spoken speech, and therefore inserts periods after the words "know" and "forever" into the speech recognition result sr11 obtained based on the spoken speech sp11 to obtain a speech recognition result sr13. On the other hand, the delimiter insertion device 10 inserts a period after the word "forever" into the speech recognition result sr21 obtained based on the spoken speech sp21 to obtain a speech recognition result sr23. The speech recognition results sr13 and sr23 have delimiters at the positions intended by the speakers of the spoken speeches sp11 and sp21.

[0031] Next, the functional units of the delimiter insertion device 10 will be described. The text acquisition unit 11 acquires target text, which is text into which delimiters are to be inserted. The target text is text obtained by speech recognition of uttered speech. The space time acquisition unit 12 acquires space times, which are the lengths of time until the next word is spoken for each word included in the uttered speech.

[0032] 4 is a diagram showing an example of acquiring target text and inter-word time. The text acquisition unit 11 acquires target text tx1 "I know it's been there forever" based on spoken speech sp3. The text acquisition unit 11 may acquire the target text by performing speech recognition of the spoken speech using well-known speech recognition processing technology or other technology.

[0033] The space time acquisition unit 12 acquires, for each word included in the target text, the length of time until the next word is spoken as the space time it. The space time acquisition unit 12 may acquire, as the space time it, a silent time during speech recognition of the spoken speech sp3. The silent time is, for example, a time during which the volume is below a predetermined level. Furthermore, the space time acquisition unit 12 may sequentially acquire a speech recognition result from a speech recognition engine used for speech recognition of the spoken speech each time speech recognition is performed, and may regard the time interval between acquisition and update of the speech recognition result as a pseudo-silence time between the occurrence of words, and acquire the update time interval as the space time it.

[0034] The delimiter insertion unit 13 inserts delimiters into the target text based on the delimiter insertion model and the inter-word time. Specifically, the delimiter insertion unit 13 inserts delimiters into the target text based on delimiter prediction information obtained by inputting the target text into the delimiter insertion model. The delimiter prediction information includes information adjusted according to the inter-word time.

[0035] The punctuation model receives as input at least punctuation-removed sentences, which are sentences that do not contain punctuation marks, and outputs punctuation prediction information indicating punctuation marks to be inserted after each word included in the punctuation-removed sentences. The punctuation model is generated by machine learning using training data that includes pairs of punctuation-removed sentences and punctuation-containing sentences, which are sentences that include punctuation marks.

[0036] Fig. 5 is a diagram showing a first example of training data used for machine learning of a delimiter insertion model. Fig. 6 is a diagram showing a first example of the configuration of a delimiter insertion model.

[0037] As shown in FIG. 5, learning data td1, which is an example of learning data used in machine learning of a delimiter insertion model md1, is made up of a pair of a delimiter-removed sentence id1 and a delimiter-included sentence od1.

[0038] The delimiter-containing sentence od1 includes a string of words that make up the sentence and delimiter labels that indicate the delimiters inserted after each word. The labels indicating the delimiters are schematically illustrated as follows in Figure 5 and other figures: <O>...no delimiter ...period, full stop <C>...comma, comma <Q>...question mark, question mark The delimiter-free sentence id1 may be a sentence obtained by removing the delimiter labels from the delimiter-containing sentence od1.

[0039] In machine learning of the delimiter insertion model md1, a delimiter-removed sentence id1 is input to the delimiter insertion model md1 in the learning process, and the weights and parameters that make up the delimiter insertion model md1 are updated based on the error between the output obtained from the delimiter insertion model md1 and the delimiter-containing sentence od1, which is the training data.

[0040] As shown in FIG. 6, the trained delimiter insertion model md1 outputs delimiter prediction information dp1 in response to input of a delimiter-removed sentence sd1.

[0041] The punctuation model md1 may be a model including a neural network. More specifically, the punctuation model md1 may be configured as a sequence labeling model that solves a sequence labeling task of predicting punctuation to be inserted after each word included in an input sentence.

[0042] The delimiter insertion model md1, which is a model including a trained neural network, can be considered as a program that is read or referenced by a computer and causes the computer to execute predetermined processes and realize predetermined functions.

[0043] That is, the trained delimiter insertion model md1 of this embodiment is used in a computer having a CPU and memory. Specifically, the CPU of the computer operates in accordance with instructions from the trained delimiter insertion model md1 stored in the memory to perform calculations on input data input to the input layer of the neural network based on trained weighting coefficients (parameters) and response functions corresponding to each layer, and to output results (probabilities) from the output layer.

[0044] The delimiter prediction information dp1 includes a symbol insertion likelihood, which is the likelihood of various delimiters that can be inserted after each word included in the delimiter-removed sentence sd1, and a symbol absence likelihood, which is the likelihood of not inserting a delimiter after each word. Then, based on the maximum likelihood of the symbol insertion likelihood and the symbol absence likelihood, it is determined (labeled) whether to insert one of multiple types of delimiters after each word or not to insert a delimiter after each word.

[0045] 6 includes the likelihood of symbol insertion and the likelihood of no symbol for each of the symbol insertions for the word "I" as follows: <O>: 90%, <C>: 5%, : 2%, <Q>: 3%. Therefore, since the likelihood of no symbol for not inserting a symbol (label <O>) is the highest, the word "I" is labeled with the label <O> for no symbol.

[0046] Similarly, the delimiter guess information dp1 includes the likelihood of symbol insertion and the likelihood of no symbol for each delimiter for the word "know" as follows: <O>: 60%, <C>: 5%, : 30%, <Q>: 5%. Therefore, since the likelihood of no symbol for not inserting a delimiter (label <O>) is the highest, the word "know" is labeled with the label <O> for no delimiter.

[0047] As an example of adjusting the delimiter prediction information based on the inter-word time, the delimiter insertion unit 13 may adjust the delimiter prediction information output from the delimiter insertion model based on the inter-word time. Specifically, the delimiter insertion unit 13 may adjust the symbol insertion likelihood and the symbol absence likelihood included in the delimiter prediction information based on the inter-word time.

[0048] The delimiter insertion unit 13 may adjust the likelihood of no symbol for one of the words included in the target text to increase when the inter-word time for the one of the words is a first time, and / or adjust the likelihood of symbol insertion for at least one of the multiple types of delimiters for the one of the words to decrease when the inter-word time for the one of the words is a second time that is longer than the first time, and may adjust the likelihood of symbol insertion for at least one of the multiple types of delimiters for the one of the words to increase and / or adjust the likelihood of no symbol for the one of the words to decrease.

[0049] Then, the delimiter insertion unit 13 inserts one of multiple types of delimiters after the word, or does not insert any delimiter, based on the maximum likelihood of the adjusted symbol insertion likelihood and the symbol-less likelihood.

[0050] To adjust the symbol insertion likelihood and the symbol absence likelihood, the delimiter insertion unit 13 may, for example, adjust the likelihood by referring to adjustment rule information. Fig. 7 is a diagram showing an example of adjustment rule information that is referred to in order to adjust the delimiter prediction information based on the inter-word time. The adjustment rule information may be stored in a storage means accessible by the delimiter insertion unit 13, or may be provided as a table inside the delimiter insertion unit 13.

[0051] 7, the adjustment rule information associates, for each range of interword time, an interword time category that indicates the length of the interword time and likelihood adjustment information that indicates the content of the likelihood adjustment. For example, the adjustment rule specifies that when the interword time (x) is 0.1 seconds or less, the interword time category is "none," and the likelihood adjustment is to "increase the likelihood of no symbol by 50%." Furthermore, the adjustment rule specifies that when the interword time (x) is longer than 0.5 seconds and less than or equal to 1.0 second, the interword time category is "medium," and the likelihood adjustment is to "increase the likelihood of symbol insertion by 50%."

[0052] 8 is a diagram showing an example of the adjustment process of delimiter prediction information. Delimiter prediction information dp21 shown in Fig. 8 shows a part of the delimiter prediction information before adjustment process output from the delimiter insertion model md1.

[0053] The delimiter prediction information dp21 includes the symbol insertion likelihood lh21 and the symbol-no-likelihood for each delimiter for the word "know." According to the delimiter prediction information dp21 before adjustment, the symbol-no-likelihood of not inserting a delimiter (label <O>) is the highest, so the word "know" is labeled with the label <O> of no delimiter.

[0054] The delimiter insertion unit 13 adjusts the likelihood of each piece of delimiter prediction information dp21 based on the space time it2 of the word "know." Specifically, since the space time it2 of the word "know" acquired by the space time acquisition unit 12 is 0.9 seconds, the delimiter insertion unit 13 refers to the adjustment rule information ( FIG. 7 ) and acquires the likelihood adjustment information "increase the symbol insertion likelihood of a period, a comma, or a question mark by 50%" associated with the space time of 0.9 seconds, and adjusts the symbol insertion likelihood lh21 of a comma, a period, or a question mark in accordance with the acquired likelihood adjustment information.

[0055] The delimiter insertion unit 13 then increases each value of the symbol insertion likelihood lh21 by 50% to obtain adjusted delimiter prediction information dp22. Since the symbol insertion likelihood of the period (label ) is the highest in the delimiter prediction information dp22, the delimiter insertion unit 13 labels the word "know" with the label of the delimiter "period". The delimiter insertion unit 13 then inserts a period after the word "know" included in the target text based on the labeled label .

[0056] In this way, the longer the inter-word time of a word included in the target text, the higher the likelihood of symbol insertion and / or the lower the likelihood of no symbol, and the shorter the inter-word time of a word, the lower the likelihood of symbol insertion and / or the higher the likelihood of no symbol, so that the speaker's intention is reflected in the likelihood of symbol insertion and the likelihood of no symbol. Then, based on the adjusted likelihood of symbol insertion and the likelihood of no symbol, a delimiter is inserted or not inserted, making it possible to obtain text in which delimiters are inserted at appropriate positions in line with the speaker's intention.

[0057] The delimiter insertion unit 13 may correct the interword times used to adjust the delimiter prediction information in accordance with predetermined conditions. Fig. 9 is a diagram showing an example of the interword time correction process. When the length of the interword times of one or more words among all words included in the target text is longer than a given value, the delimiter insertion unit 13 may correct the interword times of all words to shorten them by a given value. Then, the delimiter insertion unit 13 may adjust the symbol insertion likelihood and / or the symbol-less likelihood based on the corrected interword times, which are the corrected interword times.

[0058] 9 includes a pre-correction inter-word time it31. Here, as an example, it is assumed that the correction process is preset to reduce the inter-word times of all words by 0.5 seconds if the inter-word times of more than half of all words included in the target text are equal to or greater than the inter-word time corresponding to the inter-word time category "small."

[0059] In this case, the delimiter insertion unit 13 determines that the interword times it31 of all words included in the target text tx31 are equal to or greater than the interword time corresponding to the interword time category "small." Then, as shown in the target text tx32, the delimiter insertion unit 13 subtracts 0.5 seconds from each interword time it31 to obtain a corrected interword time it32, which is a corrected interword time. Based on the corrected interword time it32, the delimiter insertion unit 13 adjusts the symbol insertion likelihood and / or symbol-less likelihood of each word included in the target text tx32.

[0060] If a speaker's speech tends to have long inter-word times overall, there is a possibility that excessive inter-word times will be inserted in positions unintended by the speaker in the text after the insertion of the inter-word times. However, with the above-described inter-word time correction process, if the length of the inter-word time is longer than a predetermined level, the symbol insertion likelihood and / or symbol-less likelihood are adjusted based on the corrected inter-word time that has been corrected so that the inter-word time is shorter, making it possible to obtain text in which inter-word times are inserted in appropriate positions that suitably reflect the speaker's intention.

[0061] Next, a second example of the process of adjusting the punctuation prediction information according to the inter-word time will be described. Fig. 10 is a diagram showing a second example of training data used for machine learning of the punctuation insertion model. Fig. 11 is a diagram showing a second example of the configuration of the punctuation insertion model.

[0062] As shown in FIG. 10, learning data td2, which is an example of learning data used in machine learning of a delimiter insertion model md2, is made up of a pair of a delimiter-removed sentence id2 and a delimiter-included sentence od2.

[0063] The delimited sentence od2 includes a string of words that make up the sentence and delimiter labels that indicate the delimiters inserted after each word, similar to the delimiter-included sentence od1 described with reference to Fig. 5. The delimiter-removed sentence id2 includes the sentence with the delimiter labels removed from the delimiter-included sentence od2, and an inter-word time it4 associated with each word that makes up the sentence.

[0064] In machine learning of the delimiter insertion model md2, the delimiter-removed sentence id2 is input to the delimiter insertion model md2 in the learning process, and the weights and parameters that make up the delimiter insertion model md2 are updated based on the error between the output obtained from the delimiter insertion model md2 and the delimiter-containing sentence od2, which is the training data.

[0065] The punctuation model md2 may be a model including a neural network. More specifically, the punctuation model md2 may be configured as a sequence labeling model that solves a sequence labeling task of predicting punctuation to be inserted after each word included in an input sentence.

[0066] 11 , the trained punctuation model md2 outputs punctuation prediction information dp2 in response to an input of a punctuation-removed sentence sd2. The punctuation-removed sentence sd2 includes an inter-word time it5 associated with each word constituting the punctuation-removed sentence sd2. The punctuation prediction information dp2 includes a symbol insertion likelihood and a symbol absence likelihood for each word included in the punctuation-removed sentence sd2.

[0067] The delimiter insertion model md2, which is a model including a trained neural network, can be considered as a program that is read or referenced by a computer and causes the computer to execute predetermined processes and realize predetermined functions.

[0068] That is, the trained delimiter insertion model md2 of this embodiment is used in a computer having a CPU and memory. Specifically, the CPU of the computer operates in accordance with instructions from the trained delimiter insertion model md2 stored in the memory to perform calculations on input data input to the input layer of the neural network based on trained weighting coefficients (parameters) and response functions corresponding to each layer, and to output results (probabilities) from the output layer.

[0069] 6, the symbol insertion likelihood and symbol absence likelihood included in the separator prediction information dp1 output from the separator insertion model md1 are adjusted based on the space between words. In contrast, in the separator insertion model md2, the separator removal sentence id2 including space between words it4 is used as input as learning data in machine learning, and separator prediction information dp2 is output in response to the input of the separator removal sentence sd2 including space between words it5. Therefore, the symbol insertion likelihood and symbol absence likelihood included in the separator prediction information dp2 are values ​​adjusted based on the space between words by calculation in the separator insertion model md2.

[0070] In the delimiter insertion model md2, the symbol insertion likelihood and symbol absence likelihood of each delimiter for the word "I" are calculated as follows: <O>: 90%, <C>: 5%, : 2%, <Q>: 3%. Therefore, the delimiter prediction information dp2 includes information (I=<O>) indicating that the label <O> for no delimiter, which has the greatest likelihood among the symbol insertion likelihoods and symbol absence likelihoods of each delimiter for the word "I", has been labeled for the word "I". Then, the delimiter insertion unit 13 determines not to insert a delimiter after the word "I" included in the target text based on the labeled label <O>.

[0071] Furthermore, the delimiter insertion model md2 calculates the symbol insertion likelihood and symbol-absence likelihood for each delimiter for the word "know" as follows: <O>: 10%, <C>: 20%, : 70%, <Q>: 0%. In the delimiter prediction information dp1 illustrated in FIG. 6, the symbol-absence likelihood for not inserting a delimiter (labeled <O>) was the highest, whereas the likelihoods calculated by the delimiter insertion model md2 showed the symbol insertion likelihood for inserting a period (labeled ) was the highest. Therefore, the delimiter prediction information dp2 includes information (know=) indicating that the word "know" has been labeled to insert a period. The delimiter insertion unit 13 then inserts a period after the word "know" included in the target text based on the labeled label .

[0072] In this way, the delimiter insertion model md2 is generated by machine learning using, as training data, delimiter-removed sentences in which interword times are associated with each word, and the delimiter-removed sentences including the interword times of each word are used as input to the delimiter insertion model md2. Therefore, by inputting target text in which interword times are associated with each word to the delimiter insertion model, it is possible to obtain delimiter prediction information adjusted based on interword times without performing a separate adjustment process based on interword times on the output from the delimiter insertion model. Then, by inserting delimiters into the target text based on the delimiter prediction information, it is possible to easily obtain text in which delimiters are inserted in positions that conform to the speaker's intention.

[0073] Next, another example of the delimiter insertion model will be described. Fig. 12 is a diagram showing a third example of training data used for machine learning of the delimiter insertion model. In the example shown in Fig. 12, training data td3 consists of a pair of a delimiter-removed sentence id3 and a delimiter-containing sentence od3.

[0074] 5 and 10, the delimiter-containing sentence od3 includes a string of words constituting the sentence and delimiter labels indicating the delimiters inserted after each word. The delimiter-removed sentence id3 includes the sentence obtained by removing the delimiter labels from the delimiter-containing sentence od3, as well as the inter-word times associated with each word constituting the sentence, and also includes situation information st3. The situation information st3 is information indicating the situation when the speech was uttered, and may be added to the delimiter-removed sentence id3 as a tag.

[0075] The situation information st3 may have variations according to the situation when the speech voice is uttered, such as the following: Meeting: <Meeting> Lecture: <Lecture> Chatting: <Chatting>

[0076] In machine learning of the delimiter insertion model using the learning data td3, the delimiter-removed sentence id3 is input to the delimiter insertion model in the learning process, and the weights and parameters that make up the delimiter insertion model are updated based on the error between the output obtained from the delimiter insertion model and the delimiter-containing sentence od3, which is the training data.

[0077] The machine-learned punctuation model using the training data td3 outputs punctuation prediction information in response to input of a punctuation-removed sentence (target text) including situation information and inter-word times associated with each word. The output punctuation prediction information includes a likelihood of symbol insertion and a likelihood of no symbol insertion for each word included in the punctuation-removed sentence, as well as a label of whether a punctuation mark is inserted or not based on the maximum likelihood.

[0078] The delimiter insertion unit 13 determines whether to insert a delimiter according to the label after each word or not to insert a delimiter, based on delimiter prediction information obtained by inputting the target text associated with situation information into the delimiter insertion model.

[0079] In this way, a delimiter insertion model is generated by machine learning using delimiter-removed sentences associated with situation information as training data, and the delimiter-removed sentences containing situation information are used as input to the delimiter insertion model.Therefore, by inputting target text associated with situation information into the delimiter insertion model, it is possible to obtain delimiter prediction information that takes into account speech tendencies according to the situation when the spoken voice is uttered.

[0080] 1 again, output unit 14 outputs delimiter-inserted text, which is the target text into which delimiters have been inserted by delimiter insertion unit 13. The manner of output is not limited, and output unit 14 may display the delimiter-inserted text on a predetermined display means, may store the delimiter-inserted text in a predetermined storage means, or may transmit the delimiter-inserted text to a predetermined device.

[0081] 13 is a functional block diagram showing an example of the configuration of a speech recognition system according to this embodiment. As shown in FIG. 13, the speech recognition system 20 includes a delimiter insertion device 10, and is provided with a speech recognition result acquisition unit 21 and a speech recognition result output unit 22.

[0082] The speech recognition result acquisition unit 21 acquires, as a first speech recognition result, text resulting from speech recognition by a speech recognition engine and the inter-word time of each word in the text.

[0083] The text acquisition unit 11 of the delimiter insertion device 10 acquires the text included in the first speech recognition result as the target text. The space time acquisition unit 12 of the delimiter insertion device 10 acquires the space time included in the first speech recognition result.

[0084] The delimiter insertion unit 13 of the delimiter insertion device 10 performs a process of inserting delimiters using the text included in the first speech recognition result as the target text.

[0085] The speech recognition result output unit 22 outputs the target text into which the delimiters have been inserted by the delimiter insertion unit 13 as a second speech recognition result.

[0086] According to the speech recognition system 20, based on a first speech recognition result obtained from a speech recognition engine that does not take inter-word times into consideration in the speech recognition process, it is possible to obtain, with the text included in the first speech recognition result as target text, delimiter prediction information adjusted in accordance with the inter-word times included in the first speech recognition result, thereby making it possible to obtain a second speech recognition result including target text in which delimiters have been inserted at positions consistent with the speaker's intention.

[0087] FIG. 14 is a flowchart showing the processing steps of the delimiter insertion method in the delimiter insertion device 10.

[0088] In step S1, the text acquisition unit 11 acquires a target text, which is a text obtained by speech recognition of an uttered voice and is a text into which a delimiter is to be inserted.

[0089] In step S2, the space time acquisition unit 12 acquires the space time of each word in the spoken voice.

[0090] In step S3, the delimiter insertion unit 13 inputs the target text into the delimiter insertion model.

[0091] In step S4, the delimiter insertion unit 13 acquires delimiter prediction information. The delimiter prediction information includes a label indicating whether to insert any delimiter into each word or not to insert any delimiter into each word, based on the symbol insertion likelihood and symbol absence likelihood of each word adjusted according to the inter-word time.

[0092] In step S5, the delimiter insertion unit 13 inserts delimiters into the target text based on the delimiter prediction information.

[0093] In step S6, the output unit 14 outputs the delimiter-inserted text, which is the target text into which the delimiter has been inserted.

[0094] Next, a delimiter insertion program for causing a computer to function as the delimiter insertion device 10 of this embodiment will be described with reference to Fig. 15. Fig. 15 is a diagram showing the configuration of the delimiter insertion program. The delimiter insertion program P1 is composed of a main module m10 that controls the overall delimiter insertion process in the delimiter insertion device 10, a text acquisition module m11, a word space time acquisition module m12, a delimiter insertion module m13, and an output module m14. The modules m11 to m14 realize the functions of the text acquisition unit 11, word space time acquisition unit 12, delimiter insertion unit 13, and output unit 14.

[0095] The delimiter insertion program P1 may be transmitted via a transmission medium such as a communication line, or may be stored in a recording medium M1 as shown in FIG.

[0096] According to the punctuation symbol insertion device 10, punctuation symbol insertion method, and punctuation symbol insertion program P1 of the present embodiment described above, interword times in spoken speech are acquired. The interword times reflect the speaker's intention at the time of utterance. Then, punctuation symbol prediction information is acquired by inputting the target text into a punctuation symbol insertion model and adjusted according to the interword information. The acquired punctuation symbol prediction information is information adjusted according to the interword times of each word in the spoken speech, and therefore indicates punctuation symbols that reflect the speaker's intention. Then, punctuation symbols are inserted into the target text based on the punctuation symbol prediction information, making it possible to obtain text in which punctuation symbols are inserted at positions consistent with the speaker's intention.

[0097] The invention according to the present disclosure can be understood, for example, as follows.

[0098] a delimiter insertion unit that inserts delimiters that separate sentences after words included in text obtained by speech recognition of spoken speech, the delimiter insertion unit comprising: an inter-word time acquisition unit that acquires an inter-word time, which is the length of time until the next word is spoken for each word included in the spoken speech; and a delimiter insertion unit that inserts delimiters into target text, which is text obtained by speech recognition of the spoken speech, based on a delimiter insertion model and the inter-word times, the delimiter insertion model receiving as input at least a delimiter-removed sentence, which is a sentence that does not contain the delimiters, and outputting delimiter prediction information indicating delimiters to be inserted after each word included in the delimiter-removed sentence; the delimiter insertion model being a model generated by machine learning using training data including pairs of the delimiter-removed sentence and a punctuated sentence, which is a sentence that includes delimiters; and the delimiter insertion unit that inserts delimiters into the target text based on the delimiter prediction information that has been adjusted according to the inter-word time, the delimiter insertion model being obtained by inputting the target text as the delimiter-removed sentence to the delimiter insertion model.

[0099] According to the above aspect, an interword time in a spoken voice is acquired. The interword time reflects the intention of the speaker when speaking. Then, by inputting the target text into a punctuation symbol insertion model, punctuation symbol prediction information is acquired that is obtained and adjusted according to the space information. The acquired punctuation symbol prediction information is information that has been adjusted according to the interword time of each word in the spoken voice, and therefore indicates punctuation symbols that reflect the intention of the speaker. Then, by inserting punctuation symbols into the target text based on the punctuation symbol prediction information, it is possible to obtain text in which punctuation symbols are inserted at positions that conform to the intention of the speaker.

[0100] In the delimiter insertion device according to the second aspect, in the delimiter insertion device according to the first aspect, the delimiter insertion unit may adjust the delimiter prediction information output from the delimiter insertion model based on the inter-word time.

[0101] According to the above aspect, it is possible to reliably reflect the speaker's intention expressed in the inter-word time in the delimiter prediction information.

[0102] In a delimiter insertion device according to a third aspect, in the delimiter insertion device according to the second aspect, the delimiter prediction information includes a symbol insertion likelihood, which is a likelihood related to inserting each of a plurality of types of delimiters after each word included in the delimiter-removed sentence, and a no-symbol likelihood, which is a likelihood related to not inserting a delimiter after each word, and the delimiter insertion unit adjusts the no-symbol likelihood of one word to increase when the inter-word time of the one word included in the target text is a first time, and / or adjusts the no-symbol likelihood of the one word to increase when the inter-word time of the one word is a first time. The likelihood of symbol insertion may be adjusted to decrease for at least one of the plurality of types of delimiters for the one word, and when the inter-word time for the one word is a second time longer than the first time, the likelihood of symbol insertion may be adjusted to increase for at least one of the plurality of types of delimiters for the one word, and / or the likelihood of no symbol for the one word may be adjusted to decrease, and based on the maximum likelihood of the symbol insertion likelihood and the likelihood of no symbol, one of the plurality of types of delimiters may be inserted after the one word, or no delimiter may be inserted.

[0103] According to the above aspect, the longer the inter-word time of a word included in the target text, the higher the likelihood of symbol insertion and / or the lower the likelihood of no symbol, and the shorter the inter-word time of a word, the lower the likelihood of symbol insertion and / or the higher the likelihood of no symbol, thereby reflecting the speaker's intention in the likelihood of symbol insertion and the likelihood of no symbol. Then, based on the adjusted likelihood of symbol insertion and the likelihood of no symbol, a delimiter is inserted or not inserted, making it possible to obtain text in which a delimiter is inserted at an appropriate position in line with the speaker's intention.

[0104] In the delimiter insertion device according to the fourth aspect, in the delimiter insertion device according to the third aspect, the delimiter insertion unit may adjust the symbol insertion likelihood and / or the symbol-less likelihood based on a corrected inter-word time obtained by correcting the inter-word time of each of all words to be shorter by a given amount when the length of the inter-word time of one or more words among all words included in the target text is longer than a given amount.

[0105] If a speaker's speech tends to have long inter-word times overall, there is a possibility that excessive inter-word times will be inserted in positions unintended by the speaker in the text after the insertion of the inter-word times. However, according to the above aspect, when the length of the inter-word times is longer than a predetermined level, the likelihood of symbol insertion and / or the likelihood of no symbol being inserted is adjusted based on the corrected inter-word times that are corrected so that the inter-word times are shorter, making it possible to obtain text in which inter-word times are inserted in appropriate positions that suitably reflect the speaker's intention.

[0106] In a fifth aspect of the delimiter insertion device, in the delimiter insertion device according to the first aspect, the delimiter insertion model may further include, as input, an inter-word time for each word included in the delimiter-removed sentence, and may be generated by machine learning using training data consisting of pairs of the delimiter-removed sentence and the delimiter-containing sentence, in which the inter-word time is associated with each word, and may output the delimiter prediction information adjusted by the inter-word time, and the delimiter insertion unit may input the target text, in which the inter-word time is associated with each word, to the delimiter insertion model.

[0107] According to the above aspect, a punctuation mark insertion model is generated by machine learning using, as training data, punctuation mark removed sentences in which an interword time is associated with each word, and the punctuation mark removed sentences including the interword times of each word are used as input to the punctuation mark insertion model. Therefore, by inputting target text in which an interword time is associated with each word to the punctuation mark insertion model, it is possible to obtain punctuation mark prediction information adjusted based on the interword time without performing a separate adjustment process based on the interword time on the output from the punctuation mark insertion model. Then, by inserting punctuation marks into the target text based on the punctuation mark prediction information, it is possible to easily obtain text in which punctuation marks are inserted at positions that conform to the speaker's intention.

[0108] In a delimiter insertion device according to a sixth aspect, in the delimiter insertion device according to the fifth aspect, the delimiter prediction information may include a symbol insertion likelihood, which is the likelihood of inserting each of a plurality of types of delimiters after each word included in the delimiter-removed sentence, and a no-symbol likelihood, which is the likelihood of not inserting a delimiter after each word.

[0109] According to the above aspect, it is possible to obtain, for each word, punctuation prediction information including a symbol insertion likelihood and a symbol absence likelihood adjusted by the inter-word time. By inserting punctuation marks into the target text based on the adjusted symbol insertion likelihood and symbol absence likelihood, it is possible to easily obtain text in which punctuation marks are inserted at positions that conform to the speaker's intention.

[0110] In the seventh aspect of the delimiter insertion device, in the delimiter insertion device according to the fifth or sixth aspect, the delimiter insertion model may further include, as input, situation information indicating the situation when the speech corresponding to the delimiter-removed sentence was uttered, and may be generated by machine learning using training data consisting of pairs of the delimiter-removed sentence and the delimiter-containing sentence, each of which has the inter-word time associated with it and the situation information associated therewith, and the delimiter insertion unit may input the target text associated with the situation information to the delimiter insertion model.

[0111] According to the above aspect, a delimiter insertion model is generated by machine learning using delimiter-removed sentences associated with situation information as training data, and the delimiter-removed sentences including situation information are used as input to the delimiter insertion model.Therefore, by inputting target text associated with situation information into the delimiter insertion model, it is possible to obtain delimiter prediction information that takes into account speech tendencies according to the situation when the spoken voice is uttered.

[0112] A speech recognition system according to a first aspect includes a delimiter insertion device according to any one of aspects 1 to 7, and is equipped with a speech recognition result acquisition unit that acquires text resulting from speech recognition by a speech recognition engine and the inter-word times of each word in the text as a first speech recognition result, and a speech recognition result output unit, wherein the inter-word time acquisition unit of the delimiter insertion device acquires the inter-word times included in the first speech recognition result, the delimiter insertion unit of the delimiter insertion device inserts delimiters into the text included in the first speech recognition result as the target text, and the speech recognition result output unit outputs the target text into which the delimiters have been inserted by the delimiter insertion unit as a second speech recognition result.

[0113] According to the above aspect, based on a first speech recognition result obtained from a speech recognition engine that does not take inter-word times into consideration in the speech recognition process, it is possible to obtain, with the text included in the first speech recognition result as target text, delimiter prediction information adjusted in accordance with the inter-word times included in the first speech recognition result. This makes it possible to obtain a second speech recognition result including target text in which delimiters have been inserted at positions consistent with the speaker's intention.

[0114] Although the present embodiment has been described in detail above, it is clear to those skilled in the art that the present embodiment is not limited to the embodiment described in this specification. The present embodiment can be implemented in modified and altered forms without departing from the spirit and scope of the present invention as defined by the claims. Therefore, the description in this specification is intended to be illustrative and does not have any limiting meaning on the present embodiment.

[0115] The notification of information is not limited to the aspects / embodiments described in the present disclosure and may be performed using other methods. For example, the notification of information may be performed by physical layer signaling (e.g., Downlink Control Information (DCI) and Uplink Control Information (UCI)), higher layer signaling (e.g., Radio Resource Control (RRC) signaling, Medium Access Control (MAC) signaling, broadcast information (Master Information Block (MIB) and System Information Block (SIB))), other signals, or a combination thereof. Furthermore, the RRC signaling may be referred to as an RRC message, and may be, for example, an RRC Connection Setup message, an RRC Connection Reconfiguration message, or the like.

[0116] Each aspect / embodiment described herein may be applied to systems using LTE (Long Term Evolution), LTE-Advanced (LTE-A), SUPER 3G, IMT-Advanced, 4G, 5G, FRA (Future Radio Access), W-CDMA (registered trademark), GSM (registered trademark), CDMA2000, UMB (Ultra Mobile Broadband), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20, UWB (Ultra-Wide Band), Bluetooth (registered trademark), or other suitable systems and / or next generation systems enhanced therefrom. Furthermore, a combination of multiple systems (e.g., a combination of at least one of LTE and LTE-A with 5G, etc.) may also be applied.

[0117] The order of the procedures, sequences, flowcharts, etc. of each aspect / embodiment described herein may be rearranged unless it is consistent. For example, the methods described herein present elements of various steps in an example order, and are not limited to the particular order presented.

[0118] In the present disclosure, a specific operation described as being performed by a base station may be performed by its upper node in some cases. In a network consisting of one or more network nodes having a base station, it is clear that various operations performed for communication with a terminal may be performed by at least one of the base station and another network node other than the base station (for example, an MME or an S-GW, etc., but are not limited to these). Although the above example illustrates a case where there is one other network node other than the base station, a combination of multiple other network nodes (for example, an MME and an S-GW) may also be used.

[0119] Information, etc. (See the "Information, Signals" section) can be output from a higher layer (or a lower layer) to a lower layer (or a higher layer). It can also be input / output via multiple network nodes.

[0120] Input and output information may be stored in a specific location (for example, memory) or managed in a management table. Input and output information may be overwritten, updated, or added to. Output information may be deleted. Input information may be sent to another device.

[0121] The determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0122] The aspects / embodiments described in this disclosure may be used alone, in combination, or switched depending on the implementation. Notification of predetermined information (e.g., notification that "X is true") is not limited to explicit notification, but may be implicit (e.g., not notifying the predetermined information).

[0123] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is intended to be illustrative and does not have any limiting meaning on the present disclosure.

[0124] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0125] Software, instructions, etc. may also be transmitted or received over a transmission medium. For example, if the software is transmitted from a website, server, or other remote source using wired technologies such as coaxial cable, fiber optic cable, twisted pair, and Digital Subscriber Line (DSL), and / or wireless technologies such as infrared, radio, and microwave, these wired and / or wireless technologies are included within the definition of transmission media.

[0126] The information, signals, etc. described in this disclosure may be represented using any of a variety of different technologies. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0127] It should be noted that terms explained in this disclosure and / or terms necessary for understanding this specification may be replaced with terms having the same or similar meanings.

[0128] As used herein, the terms "system" and "network" are used interchangeably.

[0129] Furthermore, the information, parameters, etc. described in this specification may be expressed as absolute values, as relative values ​​from a predetermined value, or as other corresponding information. For example, a radio resource may be indicated by an index.

[0130] The names used for the above-described parameters are not intended to be limiting in any way. Furthermore, the mathematical expressions using these parameters may differ from those explicitly disclosed in this disclosure. The various channels (e.g., PUCCH, PDCCH, etc.) and information elements may be identified by any suitable names, and therefore the various names assigned to these various channels and information elements are not intended to be limiting in any way.

[0131] As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, inquiring (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may also include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in memory), and the like. Furthermore, "judgment" and "decision" can include regarding resolving, selecting, choosing, establishing, comparing, etc. as having been "judged" or "decided." In other words, "judgment" and "decision" can include regarding some action as having been "judged" or "decided." Furthermore, "judgment (decision)" can be interpreted as "assuming," "expecting," "considering," etc.

[0132] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly specified otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0133] When designations such as "first," "second," etc. are used herein, any reference to such elements does not generally limit the quantity or order of those elements. These designations may be used herein as a convenient method of distinguishing between two or more elements. Thus, a reference to a first and a second element does not imply that only two elements may be employed therein, or that the first element must precede the second element in some way.

[0134] To the extent that the terms "include," "including," and variations thereof are used herein or in the claims, these terms are intended to be inclusive, similar to the term "comprising." Furthermore, the term "or," as used herein or in the claims, is not intended to be an exclusive or.

[0135] In this disclosure, where articles are added by translation, such as a, an, and the in English, the disclosure may include that the nouns following these articles are in the plural form.

[0136] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different."

[0137] 10...delimiter insertion device, 11...target text acquisition unit, 12...space time acquisition unit, 13...delimiter insertion unit, 14...output unit, 20...speech recognition system, 21...speech recognition result acquisition unit, 22...speech recognition result output unit, M1...recording medium, m10...main module, m11...target text acquisition module, m12...space time acquisition module, m13...delimiter insertion module, m14...output module, md1, md2...symbol insertion model

Claims

1. A delimiter insertion device that inserts a delimiter for separating a sentence after a word included in text obtained by speech recognition of an utterance, comprising: an inter-word time acquisition unit that acquires an inter-word time, which is a length of time until the next word is uttered for each word included in the uttered voice; a delimiter insertion unit that inserts delimiters into a target text, the target text being a text obtained by speech recognition of the spoken voice, based on a delimiter insertion model and the inter-word time, The delimiter insertion model is At least a delimiter-removed sentence, which is a sentence that does not include the delimiter, is input; outputting punctuation information indicating a punctuation mark to be inserted after each word included in the punctuation mark removed sentence; a model generated by machine learning using learning data including pairs of the punctuation-removed sentences and punctuation-containing sentences, the punctuation-removed sentences being sentences that include punctuation marks, a delimiter insertion unit that inserts delimiters into the target text based on the delimiter prediction information obtained by inputting the target text as the delimiter-removed sentence into the delimiter insertion model and adjusted according to the inter-word time; A delimiter insertion device comprising:

2. The punctuation symbol insertion unit adjusts the punctuation symbol prediction information output from the punctuation symbol insertion model based on the inter-word time.

2. The delimiter insertion device of claim 1.

3. the delimiter prediction information includes a symbol insertion likelihood, which is a likelihood regarding the insertion of each of a plurality of types of delimiters after each word included in the delimiter-removed sentence, and a symbol absence likelihood, which is a likelihood regarding the non-insertion of a delimiter after each word; The delimiter insertion unit is When the inter-word time in one word among the words included in the target text is a first time, adjusting the symbol-free likelihood of the one word to be increased, and / or adjusting the symbol insertion likelihood of at least one delimiter of the one word among a plurality of types of delimiters to be decreased; when the inter-word time in the one word is a second time longer than the first time, adjusting the symbol insertion likelihood of at least one of a plurality of types of delimiters of the one word so as to increase, and / or adjusting the symbol absence likelihood of the one word so as to decrease, inserting any one of a plurality of types of delimiters after the one word, or not inserting any delimiter, based on the maximum likelihood of the symbol insertion likelihood and the symbol absence likelihood; 3. The delimiter insertion device according to claim 2.

4. The delimiter insertion unit is when the extent of the length of the inter-word time of one or more words among all words included in the target text is longer than a given extent, adjusting the symbol insertion likelihood and / or the symbol absence likelihood based on a corrected inter-word time obtained by correcting the inter-word time of each of all words so as to be shorter by a given extent; 4. The delimiter insertion device according to claim 3.

5. The delimiter insertion model is further comprising as input an inter-word time for each word in the delimiter removed sentence; the inter-word times are generated by machine learning using learning data consisting of pairs of the punctuation-removed sentences and the punctuation-containing sentences associated with each word, outputting the punctuation guess information adjusted by the inter-word time; The delimiter insertion unit inputs the target text, in which each word is associated with the inter-word time, into the delimiter insertion model.

2. The delimiter insertion device of claim 1.

6. the delimiter prediction information includes a symbol insertion likelihood, which is a likelihood regarding the insertion of each of a plurality of types of delimiters after each word included in the delimiter-removed sentence, and a symbol absence likelihood, which is a likelihood regarding the non-insertion of a delimiter after each word; 6. A delimiter insertion device according to claim 5.

7. The delimiter insertion model is The speech recognition system further includes, as an input, situation information indicating a situation when the speech corresponding to the delimiter-removed sentence was uttered, the inter-word time is associated with each word and the situation information is associated with the sentences, and the learning data is generated by machine learning using learning data consisting of pairs of the sentences with and without punctuation and the sentences with and without punctuation, The delimiter insertion unit inputs the target text associated with the situation information into the delimiter insertion model.

6. The delimiter insertion device according to claim 5.

8. A speech recognition system including the delimiter insertion device according to claim 1, a speech recognition result acquisition unit that acquires, as a first speech recognition result, a text resulting from speech recognition by a speech recognition engine and the inter-word time of each word in the text; A speech recognition result output unit; the space time acquisition unit of the delimiter insertion device acquires the space time included in the first speech recognition result, The delimiter insertion unit of the delimiter insertion device inserts delimiters into the text included in the first speech recognition result as the target text; the speech recognition result output unit outputs, as a second speech recognition result, the target text in which the delimiter has been inserted by the delimiter insertion unit. A voice recognition system comprising: