JP2010266716A - Voice recognition device, and method and program of the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To convert input voice into a recognition result easily understood by a user and output it. <P>SOLUTION: In this voice recognition device, phoneme information as an acoustic feature of each phoneme is previously stored in an acoustic model storage section. A word, reading way of this word, difficulty level of this word, and a word having the same meaning as this word and different difficulty level from it are previously stored in association with each other in a recognition dictionary storage section. A linguistic feature is previously stored in a language model storage section. The acoustic feature of voice information is determined using the voice information, similar phoneme information is retrieved from the acoustic model storage section using the acoustic feature of the voice information determined by a voice analysis section, and a word is retrieved from the recognition dictionary storage section using one or more retrieved pieces of phoneme information. Using the retrieved word and the linguistic feature of a language model storage section, a recognition result before conversion is estimated, the recognition dictionary storage section is referred, and a word of a difficulty level closest to a target difficulty level is selected from words having the same meaning as the word of the recognition result before conversion. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus, a speech recognition method, and a speech recognition program that recognize the content of speech information using an acoustic model and a language model and obtain the content as text data.

  An apparatus for performing speech recognition using an acoustic model and a language model is known as a conventional technique. FIG. 1 shows a configuration example of a conventional speech recognition apparatus 20. The voice recognition device 20 receives voice through the voice input unit 21. The speech analysis unit 23 analyzes the input speech, and the search unit 25 uses the acoustic model storage unit 27 and the language model storage unit 31 to estimate what word sequence appears, and as a result of the estimation, The word sequence with the highest probability is output as the speech recognition result. At this time, in the word sequence estimation using the language model storage unit 31, the words registered in the recognition dictionary storage unit 29 are arranged to form a word sequence.

  Patent Document 1 is known as a method for making various recognition words visible as the same output characters by associating the same output characters with various recognition words in a list.

Japanese Patent Laid-Open No. 2005-309065

  In the prior art, when all words corresponding to the spoken sentence are registered in the recognition dictionary storage unit and speech recognition works ideally, it is recognized if a difficult word (for example, “regret”) is spoken. The result is also difficult text data. Since the purpose of the conventional speech recognition is “to make the uttered content exactly as it is”, there is no problem even if the recognition result is difficult text data. However, when the purpose is to "text the spoken content so that the user can easily understand", there is a problem that the input speech is converted into a recognition result that is easy for the user to understand and output. .

  In order to solve the above problem, in the speech recognition technology of the present invention, phoneme information that is an acoustic feature of each phoneme is stored in advance in the acoustic model storage unit, and a word and the word are stored in the recognition dictionary storage unit. , The difficulty level of this word, and a word having the same meaning as this word and having a different difficulty level are stored in advance, and language features are stored in advance in the language model storage unit. The acoustic feature of the speech information is obtained using speech information, and similar phoneme information is searched from the acoustic model storage unit using the acoustic feature of the speech information obtained by the speech analysis unit. The phoneme information is used to search for a word from the recognition dictionary storage unit, and the one or more searched words and the linguistic features of the language model storage unit are used to estimate the recognition result before conversion, and the recognition dictionary storage unit is Refer to the same word as the recognition result before conversion. It means to select the closest word to the degree of difficulty of the target from within a word with.

  The present invention provides an effect that it is possible to obtain a recognition result that is easy for the user to understand by providing the conversion unit.

The figure which shows the structural example of the conventional speech recognition apparatus 20. FIG. The figure which shows the structural example of the speech recognition apparatus. The figure which shows the example of a processing flow of the speech recognition apparatus 100. The figure which shows the example of data memorize | stored in the recognition dictionary memory | storage part 129. The figure which shows the example of a processing flow of the conversion part. (A) is an example of a recognition result before conversion including a word with difficulty level 3, and (B) is an example of a recognition result after conversion when difficulty level 2 is set as difficulty level selection information with (A) being a recognition result before conversion. , (C) is a recognition result example after conversion when (A) is the recognition result before conversion and difficulty level 1 is set as the difficulty level selection information. (D) is a difficulty level selection with (A) being the recognition result before conversion. Example of recognition result after conversion when difficulty level 1 is set as information and non-displayable is set as display selection information, (E) is an example of recognition result before conversion including a word of difficulty level 1, (F) is The figure which shows the example of a recognition result after conversion when the difficulty level 2 is set as the recognition result before conversion as (E) and difficulty level selection information. The block diagram which illustrated the hardware constitutions of the speech recognition apparatus 100 in a present Example. FIG. 10 is a diagram illustrating a configuration example of a conversion unit 210 according to the second embodiment. The figure which shows the example of a processing flow of the conversion part 210.

  Hereinafter, embodiments of the present invention will be described in detail.

<Voice recognition apparatus 100>
FIG. 2 shows a configuration example of the speech recognition apparatus 100, and FIG. 3 shows a processing flow example of the speech recognition apparatus 100. The speech recognition apparatus 100 according to the first embodiment is described with reference to FIGS. 2 and 3. The speech recognition apparatus 100 recognizes the content of speech information and obtains the content as text data. The speech recognition apparatus 100 includes, for example, a language model 131 including a speech input unit 21, a speech analysis unit 23, a search unit 25, an acoustic model storage unit 27, and a recognition dictionary storage unit 129, a conversion unit 110, a storage unit 103, and a control unit 105. Have The obtained text data may be output to an output device (not shown) (for example, a display or a printer), an external recording medium, or the like, or may be stored in the storage unit 103 described later.

<Storage unit 103 and control unit 105>
The storage unit 103 stores / reads each input / output data and each data of the calculation process one by one. Thereby, each calculation process is advanced. However, the data need not necessarily be stored in the storage unit 103, and data may be directly transferred between the units.
The control unit 105 controls each process.

<Voice input unit 21>
The sound input unit 21 is, for example, a microphone and an input interface, and converts sound into an electrical signal, and further includes an A / D converter and converts the sound into digital data. In this specification, voice, voice converted to an electrical signal, and voice digital data converted by an A / D converter are collectively referred to as voice information. The voice recognition apparatus 100 receives a voice via the voice input unit 21 (s21). However, when recognizing the content of the audio information stored in the external recording medium or the storage unit 103, the audio input unit 21 may not be provided.

<Speech analysis unit 23>
The voice analysis unit 23 obtains an acoustic feature of the voice information using the voice information (s23). For example, the acoustic feature is MFCC (mel-frequency cepstral coefficient) or the like.

<Acoustic model storage unit 27>
The acoustic model storage unit 27 stores phoneme information that is an acoustic feature of each phoneme in advance. For example, the acoustic model storage unit 27 stores a MFCC of a standard pattern (standard model) or the like in advance.

<Recognition dictionary storage unit 129>
FIG. 4 shows an example of data stored in the recognition dictionary storage unit 129. The recognition dictionary storage unit 129 stores a word, a reading of the word, a difficulty level of the word, and a word having the same meaning as the word and a difficulty level different from each other. In addition, for example, the recognition dictionary storage unit 129 may store a word in advance in association with a display flag that determines whether to display the word after conversion. For example, data stored in the recognition dictionary storage unit 129 is registered in advance by a manufacturer, a user, or the like prior to use.

<Language model storage unit 131>
The language model storage unit 131 stores linguistic features in advance. For example, the language model storage unit includes a recognition dictionary storage unit 129, and the occurrence probability of each word string is stored in advance.

<Search unit 25>
The search unit 25 searches for similar phoneme information from the acoustic model storage unit using the acoustic features of the voice information obtained by the voice analysis unit 23. Further, the search unit 25 searches for a word from the recognition dictionary storage unit 129 using the searched one or more phoneme information. Finally, the search unit 25 estimates the recognition result before conversion using one or more searched words and the linguistic features of the language model storage unit 131 (s23).

  For example, the search unit 25 calculates Euclidean distance from the acoustic feature (MFCC) obtained by the speech analysis unit 23 and the MFCC of the standard pattern stored in the acoustic model storage unit 27, and searches for similar phoneme information. Furthermore, reading is estimated from the searched phoneme information using a hidden Markov model, and a corresponding word is searched from the recognition dictionary storage unit 129. A word string having a high probability is searched from a word string composed of the searched words and the occurrence probability of the word string stored in the language model storage unit 131, and a recognition result before conversion is estimated.

  Note that speech recognition generally refers to obtaining the left side P (w ^ | y) of the following equation, where w is the text data to be estimated and y is the time-series sound obtained by the speech analysis unit 23. P (y | w) represents information held by the acoustic model storage unit 27, and P (w) represents information held by the language model storage unit 131 (reference: written by Seiichi Nakagawa, “By Probability Model”). Speech recognition ", Corona, July 1, 1988, p.33-p.34).

  However, the present invention is not limited to this embodiment. For example, the processing contents of the voice input unit 21, the voice analysis unit 23, the search unit 25, and the acoustic model storage unit 27 may calculate the recognition result before conversion using another conventional technique. The recognition dictionary storage unit 129 and the language model storage unit 131 may store and use information (for example, class of each word) used in the prior art in addition to the above information.

<Conversion unit 110>
FIG. 5 shows a processing flow example of the conversion unit 110. The conversion unit 110 refers to the recognition dictionary storage unit 129 and selects a word closest to the target difficulty level from words having the same meaning as the word of the recognition result before conversion (s110). The word having the same meaning is a word associated with the word itself and the recognition dictionary storage 129.

  For example, the conversion unit 110 selects a target difficulty level after conversion in advance, refers to the recognition dictionary storage unit 129, and has the same meaning as the word of the recognition result before conversion and is closest to the selected difficulty level. Is selected (s116). Further, the conversion unit 110 refers to the recognition dictionary storage unit 129 to determine whether or not to display the word of the recognition result before conversion after conversion (s115), and when the display flag indicates that it is displayed. When the word closest to the target difficulty level is selected from the words having the same meaning as this word (s116) and it means that the display flag is not displayed, the converted recognition result is the same as this word. Conversion is performed so that a meaningful word is not displayed (s117).

<Example of processing flow>
For example, the user transmits setting information such as difficulty level selection information and display selection information to the conversion unit 110 by using an input device (for example, a mouse, a keyboard, etc.) not shown in advance. The difficulty level selection information is information that determines the difficulty level of the word that is the target of the recognition result after conversion. For example, the difficulty level selection information is set to 1. The display selection information is information for determining whether or not to display a word corresponding to the recognition result after conversion, and is set to be non-displayable or non-displayable, for example. However, when the target difficulty level of the speech recognition apparatus 100 is determined in advance (for example, “difficulty level 2”), whether or not to display words having the same meaning in the post-conversion recognition result is determined in advance. In some cases (for example, “non-display not possible”), it is not necessary to input the difficulty level selection information and the display selection information.

  The conversion unit 110 receives the recognition result before conversion and outputs the recognition result after conversion. The conversion unit 110 stores the recognition result before conversion input from the search unit 25 in a buffer or the like, and extracts one word therefrom (s112). If the display selection information is set to be non-displayable, it is determined whether or not the display flag of the extracted word is non-displayable with reference to the recognition dictionary 129 (s115), and the display flag cannot be non-displayable. In the case of, a word closest to the target difficulty level is selected from words having the same meaning as this word (s116). When the word itself is selected as a word having the same meaning in s116, the conversion unit 110 records the word in an output buffer or the like without performing any conversion process, and the word has a different difficulty level. When is selected, the word is converted into the selected word, and the converted word is recorded in an output buffer or the like. If the display flag is non-displayable, conversion is performed so that this word is not displayed in the recognition result after conversion (s117). For example, this word is deleted. If the display selection information is set to be non-displayable, the display flag determination (s115) and non-display conversion (s117) are not performed, and the conversion process (s116) is performed for each word.

  The word subjected to the conversion process is stored in an output buffer or the like (not shown). It is determined whether or not it is the last word of the recognition result before conversion (s118). If it is not the last word, one next word is extracted from the recognition result before conversion and the above process is repeated. In the case of the last word, as a recognition result after conversion, the recognition result obtained by the above conversion is extracted from an output buffer or the like and output (s119). However, the configuration may be such that the recognition result after conversion is output for each conversion. In this case, after outputting the recognition result (s119), it is determined whether or not it is the last word (s118).

<Specific example>
FIG. 6A shows an example of a recognition result before conversion including a word with difficulty level 3, and FIG. 6B shows a recognition result after conversion when difficulty level selection information is set as difficulty recognition information with (A) as a recognition result before conversion. For example, (C) is a recognition result example after conversion when (A) is a recognition result before conversion and difficulty level 1 is set as difficulty level selection information, and (D) is a difficulty with (A) being a recognition result before conversion. An example of a recognition result after conversion when difficulty level 1 is set as degree selection information and non-displayable is set as display selection information, (E) is an example of recognition result before conversion including a word of difficulty level (F) ) Shows an example of a recognition result after conversion when (E) is a recognition result before conversion and difficulty level 2 is set as difficulty level selection information. In the conversion processes (B), (C), and (F), the display selection information is set to be non-displayable. In (A), the difficulty level is displayed at the bottom of each word for ease of explanation, but it may not be displayed in actual use. The underline of the word to be converted may not be displayed in the same manner.

  For example, as shown in the data example of FIG. 4, when information is registered in advance in the recognition dictionary storage unit 129 and the difficulty level selection information is difficulty level 1, the recognition results before conversion include “today” and “I”. When difficulty level 2 is input, it is converted to “Kyo” and “I” with difficulty level 1 and output. When difficulty level selection information is difficulty level 2, if difficulty level 3 such as “I am” or “extreme pain” is input as the recognition result before conversion, “I”, “very” Convert to "Sorry" and output. In this way, for example, when the pre-conversion recognition result of FIG. 6A is input to the conversion unit 110, the post-conversion recognition result of FIG. 6B is output if the difficulty level 2 is set. If the difficulty level 1 is set, the recognition result after conversion shown in FIG. 6C is output. In FIG. 6A, “Ut” is difficulty level 1. However, in FIG. 4, there are no words having the same meaning with different difficulty levels, so the word having the same meaning closest to the selected difficulty level 2 is the word “Uto” itself, and is not subjected to conversion processing, and is used for output. Is output to the buffer. Further, in FIG. 6A, “will of regret” has a difficulty level of 3. When the difficulty level indicated by the difficulty level selection information is 1, since the word of difficulty level 1 corresponding to “regret” is not registered in FIG. 4, the word “sorry feeling” corresponding to difficulty level 2 is not registered. "Is the word having the same meaning closest to the selected difficulty level 1.

  In addition, it may be difficult to read any word when all spoken words are displayed. For example, an adverb that indicates the degree of “very” or the like is a word that can be understood even if it is not described. In some cases, it may be easier to understand the recognition result if it is not displayed in the display result after conversion of an ambiguous word such as “um” or “an” or an adverb indicating the degree of “very” or “just”. In this embodiment, when a “display flag” is registered in advance for each word in the recognition dictionary storage unit 129 and it means that the display flag is not displayed (can be hidden), the recognition result after conversion is the same as the word. Convert so that words with meaning are not displayed. In this case, nothing needs to be output to the output buffer or the like. When the pre-conversion recognition result of FIG. 6 (A) is input to the conversion unit 10 and difficulty level 1 is selected as difficulty level selection information and non-display is selected as display selection information, after the conversion of FIG. 6 (D). Output the recognition result. In this case, the conversion is made so that the word “um” is not displayed.

  When the difficulty level selection information is difficulty level 2, when difficulty level 1 such as “Anyo”, “Natachatte”, “Mama”, etc. is input as the recognition result before conversion, It can be converted into “leg”, “get”, “mom” and output. For example, when FIG. 6E is input and the difficulty level 2 is selected as the difficulty level selection information, the recognition result after conversion in FIG. 6F is output. By adopting such a configuration, it is possible to convert an infant word or the like having a low difficulty level into a recognition result before conversion having a high difficulty level, and a recognition result that is easy for the user to understand can be obtained.

  In addition, when the words “I” and “I” with the same meaning are included in one utterance content (pre-conversion recognition result), it is converted into the same word and the recognition result is easy for the user to understand. Can be requested.

  In the present embodiment, the difficulty level is set to three levels from 1 to 3. However, the difficulty level does not have to be three levels, and may be two levels or four or more levels.

  The conversion unit 110 may be a part of the search unit 25. In this case, the search unit 25 performs a conversion process each time a pre-conversion recognition result word is estimated. In this case, the process of extracting one word from the pre-conversion recognition result in FIG. 5 (s112) and the determination of whether or not it is the last word (s118) may not be performed.

<Effect>
With such a configuration, it is possible to obtain a recognition result that is easy for the user to understand. By unifying the difficulty level of each word in the text data of the recognition result, the user can easily understand. For example, a child or a foreigner who is not good at Japanese can reduce the difficulty level and present the recognition result in plain expression. Conversely, when the user has a high language ability with respect to the speaker, a recognition result having a high difficulty level can be presented. By inputting the difficulty level selection information to the conversion unit 110, text data with different difficulty levels can be presented according to the user's understanding level. By providing a display flag for data registered in the recognition dictionary storage unit 129 and inputting display selection information to the conversion unit 110, words with low importance can be hidden as necessary, and text data can be displayed. It can be presented more clearly and concisely. In addition, in the creation of minutes, etc., that are specified to use specific terms (words), the terms can be presented in a unified manner.

<Hardware configuration>
FIG. 7 is a block diagram illustrating a hardware configuration of the speech recognition apparatus 100 according to the present embodiment. As illustrated in FIG. 7, the speech recognition apparatus 100 of this example includes a CPU (Central Processing Unit) 11, an input unit 12, an output unit 13, an auxiliary storage device 14, a ROM (Read Only Memory) 15, and a RAM (Random). Access Memory) 16 and a bus 17.

  The CPU 11 in this example includes a control unit 11a, a calculation unit 11b, and a register 11c, and executes various calculation processes according to various programs read into the register 11c. The input unit 12 is an input interface for inputting data, a keyboard, a mouse, and the like, and the output unit 13 is an output interface for outputting data. The auxiliary storage device 14 is, for example, a hard disk, an MO (Magneto-Optical disc), a semiconductor memory, or the like, and stores a program area 14a in which a program for causing the computer to function as the voice recognition device 100 is stored and various data. It has a data area 14b. The RAM 16 is an SRAM (Static Random Access Memory), a DRAM (Dynamic Random Access Memory), or the like, and has a program area 16a in which the above programs are stored and a data area 16b in which various data are stored. The bus 17 connects the CPU 11, the input unit 12, the output unit 13, the auxiliary storage device 14, the ROM 15, and the RAM 16 so that they can communicate with each other. In addition, as a specific example of such hardware, a server apparatus, a workstation, etc. other than a personal computer can be illustrated, for example.

<Program structure>
As described above, each program for executing each process of the speech recognition apparatus 100 of the present embodiment is stored in the program areas 14a and 16a. Each program constituting the speech recognition program may be described as a single program sequence, or at least a part of the program may be stored in the library as a separate module. In addition, each program may realize each function alone, or each program may read each other library to realize each function.

<Cooperation between hardware and program>
The CPU 11 (FIG. 7) writes the above-described program stored in the program area 14 a of the auxiliary storage device 14 in the program area 16 a of the RAM 16 in accordance with the read OS (Operating System) program. Similarly, the CPU 11 writes various data stored in the data area 14 b of the auxiliary storage device 14 in the data area 16 b of the RAM 16. The address on the RAM 16 where the program and data are written is stored in the register 11c of the CPU 11. The control unit 11a of the CPU 11 sequentially reads these addresses stored in the register 11c, reads a program and data from the area on the RAM 16 indicated by the read address, causes the calculation unit 11b to sequentially execute the operation indicated by the program, The calculation result is stored in the register 11c.

  FIG. 2 is a block diagram illustrating the functional configuration of the speech recognition apparatus 100 configured by reading and executing the above-described program in the CPU 11 as described above.

  Here, the storage unit 103 corresponds to any one of the auxiliary storage device 14, the RAM 16, the register 11 c, other buffer memory and cache memory, or a storage area using these in combination. In addition, the storage unit 103, the control unit 105, the speech analysis unit 23, the acoustic model storage unit 27, the recognition dictionary storage unit 129, the language model storage unit 131, the search unit 25, and the conversion unit 110 cause the CPU 11 to execute a speech recognition program. It is constituted by. Further, the speech recognition apparatus 100 according to the present embodiment executes each process under the control of the control unit 105.

Only parts different from the first embodiment will be described. The configuration of the conversion unit is different from that of the first embodiment.
<Conversion unit 210>
FIG. 8 illustrates a configuration example of the conversion unit 210 according to the second embodiment, and FIG. 9 illustrates a processing flow example of the conversion unit 210. The conversion unit 210 includes a processing unit 211 and a representative value calculation unit 217.

<Representative value calculation unit 217>
The representative value calculation unit 217 obtains a representative value of the difficulty level in a predetermined section of the recognition result before conversion (s213).

<Processing unit 211>
The processing unit 211 uses the representative value obtained as the target difficulty level, refers to the recognition dictionary storage unit 129, and selects the word that has the same meaning as the word of the recognition result before conversion and is closest to the representative value (s216). .

<Example of processing flow>
The converter 210 receives the pre-conversion recognition result and outputs the post-conversion recognition result. The conversion unit 210 stores the recognition result before conversion input from the search unit 25 in a buffer or the like. Then, words for a predetermined section are extracted, the difficulty level of each word is acquired with reference to the recognition dictionary 129, and a representative value of the difficulty level of the word in the predetermined section is calculated (s213). This is performed for each section, and representative values for all sections are calculated. Note that the predetermined section may be the number of words to be converted, the processing time, or all the words of the recognition result before conversion. As the representative value, for example, an average value, a mode value, a median value, or the like can be considered.

  Next, one word of the recognition result before conversion is extracted from the buffer or the like (s112). If the display selection information is set to be non-displayable, it is determined whether or not the display flag of the extracted word is non-displayable with reference to the recognition dictionary 129 (s115), and the display flag cannot be non-displayable. In the case of, a word closest to the difficulty level of the representative value is selected from words having the same meaning as this word (s216). When the word itself is selected as a word having the same meaning in s216, the conversion unit 110 records the word in an output buffer or the like without performing a conversion process, and the word has a different difficulty level. When is selected, the word is converted into the selected word, and the converted word is recorded in an output buffer or the like. If the display flag is non-displayable, conversion is performed so that this word is not displayed in the recognition result after conversion (s117). Each word is converted into a word closest to the difficulty level of the representative value of the section to which each word belongs. If the display selection information is set to be non-displayable, the display flag determination (s115) and non-display conversion (s117) are not performed, and conversion processing (s216) is performed for each word.

  The word subjected to the conversion process is stored in an output buffer or the like (not shown). It is determined whether or not it is the last word of the recognition result before conversion (s118). If it is not the last word, one next word is extracted from the recognition result before conversion and the above process is repeated. In the case of the last word, as a recognition result after conversion, the recognition result obtained by the above conversion is extracted from an output buffer or the like and output (s119). However, the configuration may be such that the recognition result after conversion is output for each conversion.

  For example, when the recognition result before conversion shown in FIG. 6A is input, assuming that a predetermined section is all words of the recognition result before conversion and the representative value is an average value, the representative value of the difficulty of the predetermined section is 1.5. For example, when this value is rounded up and the representative value of the difficulty level is 2, the recognition result after conversion is as shown in FIG.

<Effect>
With such a configuration, it is possible to obtain a recognition result that is easy for the user to understand. For example, by obtaining a representative value (for example, an average value) of the difficulty level of a predetermined section (for example, the entire recognition result before conversion) from the difficulty level of each word of the recognition result before conversion, the entire predetermined section including each word is obtained. If a difficulty level trend is required, and these representative values are used as the difficulty level of the recognition result after conversion, words with difficulty levels that are extremely different from the surrounding words are prevented from being displayed as they are, and difficulty levels that match the context The word of the degree can be displayed.

  In the present invention, the recognition result before conversion is not simply converted into words having the same meaning, but is converted using the degree of difficulty, so that it is possible to obtain text data that is easy for the user to understand.

  In addition, as setting information, it is set as the structure which inputs the conversion setting information which selects either the method (manual) of Example 1 and the method (automatic) of Example 2 to a conversion part, and Example 1 and Example 2 are combined. Can also be used. In this case, the user can select text data to be output. When manual is selected, difficulty level selection information is input to the conversion unit.

DESCRIPTION OF SYMBOLS 100 Speech recognition apparatus 21 Speech input part 23 Speech analysis part 25 Search part 27 Acoustic model memory | storage part 129 Recognition dictionary memory | storage part 131 Language model memory | storage part 110,210 Conversion part 211 Processing part 217 Representative value calculation part

Claims (9)

A speech recognition device that recognizes the content of speech information and obtains the content as text data,
A voice analysis unit for obtaining acoustic characteristics of the voice information using the voice information;
An acoustic model storage unit in which phoneme information that is an acoustic feature of each phoneme is stored in advance;
A recognition dictionary storage unit which stores in advance a word, a reading of the word, a difficulty level of the word, and a word having the same meaning as the word and a difficulty level different from each other;
A language model storage unit in which linguistic features are stored in advance;
Search for similar phoneme information from the acoustic model storage unit using the acoustic features of the speech information obtained by the speech analysis unit, and search for words from the recognition dictionary storage unit using the searched one or more phoneme information. A search unit that estimates a recognition result before conversion using one or more searched words and linguistic features of the language model storage unit;
With reference to the recognition dictionary storage unit, a conversion unit that selects a word closest to the target difficulty level from words having the same meaning as the word of the recognition result before conversion,
A speech recognition apparatus. The speech recognition apparatus according to claim 1,
The conversion unit selects a target difficulty level after conversion in advance, refers to the recognition dictionary storage unit, and selects a word closest to the selected difficulty level with the same meaning as the word of the recognition result before conversion. select,
A speech recognition apparatus characterized by that. The speech recognition apparatus according to claim 1,
The conversion unit includes a representative value calculation unit that obtains a representative value of the difficulty level in a predetermined section of the recognition result before conversion, sets the target difficulty level as the representative value, and refers to the recognition dictionary storage unit, Selecting a word that has the same meaning as the word of the recognition result before conversion and is closest to the representative value;
A speech recognition apparatus characterized by that. The speech recognition device according to any one of claims 1 to 3,
The recognition dictionary storage unit determines a word, a reading of the word, a difficulty level of the word, a word having the same meaning as the word and a different difficulty level, and whether to display the word after conversion. Pre-stored in association with the display flag,
The conversion unit refers to the recognition dictionary storage unit, determines whether or not to display the word of the recognition result before conversion after conversion, and displays the display flag of the word of the recognition result before conversion If it means that the word closest to the target difficulty level is selected from the words having the same meaning as the word, and the display flag is not displayed, Convert to not display words with the same meaning,
A speech recognition apparatus characterized by that. A speech recognition method for recognizing the content of speech information using an acoustic model storage unit, a recognition dictionary storage unit, and a language model storage unit, and obtaining the content as text data,
The phone model information that is the acoustic feature of each phoneme is stored in advance in the acoustic model storage unit, and the word, the reading of the word, the difficulty of the word, and the same as the word are stored in the recognition dictionary storage unit Assume that words having different meanings and different difficulty levels are stored in advance, and language features are stored in advance in the language model storage unit.
A voice analysis step for obtaining an acoustic feature of the voice information using the voice information;
Search for similar phoneme information from the acoustic model storage unit using the acoustic features of the speech information obtained in the speech analysis step, and search for words from the recognition dictionary storage unit using the searched one or more phoneme information. A search step for estimating a pre-conversion recognition result using one or more searched words and linguistic features of the language model storage unit;
A conversion step of referring to the recognition dictionary storage unit and selecting a word closest to a target difficulty level from words having the same meaning as the word of the recognition result before conversion;
A speech recognition method comprising: The speech recognition method according to claim 5,
In the conversion step, a target difficulty level after conversion is selected in advance, and referring to the recognition dictionary storage unit, a word that has the same meaning as the word of the recognition result before conversion and is closest to the selected difficulty level is selected. select,
A speech recognition method characterized by the above. The speech recognition method according to claim 5,
The conversion step includes a representative value calculation step for obtaining a representative value of the difficulty level in a predetermined section of the recognition result before conversion, the target difficulty level as the representative value, and referring to the recognition dictionary storage unit, Selecting a word that has the same meaning as the word of the recognition result before conversion and is closest to the representative value;
A speech recognition method characterized by the above. The speech recognition method according to any one of claims 5 to 7,
The recognition dictionary storage unit determines a word, a reading of the word, a difficulty level of the word, a word having the same meaning as the word and a different difficulty level, and whether to display the word after conversion. Pre-stored in association with the display flag,
The conversion step refers to the recognition dictionary storage unit to determine whether to display the pre-conversion recognition result word after conversion, and to display the pre-conversion recognition word display flag If it means that the most vowed word is selected for the target difficulty level from words having the same meaning as that of the word, and the display flag is not displayed, the word is included in the recognition result after conversion. To avoid displaying words with the same meaning as
A speech recognition method characterized by the above.   A speech recognition program for causing a computer to function as the speech recognition apparatus according to claim 1.

Images