JP2020034832A - Dictionary generation device, voice recognition system, and dictionary generation method - Google Patents

Abstract

To improve recognition rate of voice recognition.SOLUTION: A dictionary generation device according to an embodiment includes a receiving unit, a generation unit, and an update unit. The receiving unit receives an uttered voice of a word with a pronunciation different from the standard pronunciation. The generation unit generates a phoneme sequence based on the uttered voice received by the receiving unit. The update unit updates a voice dictionary by adding the phoneme sequence generated by the generation unit to the voice dictionary.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a dictionary generation device, a speech recognition system, and a dictionary generation method.

  2. Description of the Related Art Conventionally, for example, there is an in-vehicle device that receives a voice operation by a user's speech. In such an in-vehicle device, it is common to accept a voice operation by searching a dictionary for voice recognition based on an uttered voice.

  Further, Patent Document 1 discloses that when there are a plurality of candidates based on the uttered voice, an additional voice input is accepted to finally narrow down one candidate.

JP-A-11-250078

  However, in the prior art, there is room for improvement in improving the recognition rate of speech recognition. That is, since a generally used speech dictionary for speech recognition is created based on native utterances, the recognition rate is lower for non-native utterances.

  The present invention has been made in view of the above, and an object of the present invention is to provide a dictionary generation device, a voice recognition system, and a dictionary generation method that can improve the recognition rate of voice recognition.

  In order to solve the above-described problem and achieve the object, a dictionary generation device according to an embodiment includes a reception unit, a generation unit, and an update unit. The receiving unit receives an uttered voice of a word having a pronunciation different from a standard pronunciation. The generating unit generates a phoneme sequence based on the uttered voice received by the receiving unit. The updating unit updates the voice dictionary by adding the phoneme string generated by the generating unit to the voice dictionary.

  According to the present invention, the recognition rate of voice recognition can be improved.

FIG. 1 is a diagram showing an outline of a dictionary generation method. FIG. 2 is a schematic diagram of the speech recognition system. FIG. 3 is a block diagram of the dictionary generation device. FIG. 4 is a diagram illustrating a specific example of the speech dictionary. FIG. 5 is a diagram illustrating an example of the rule information. FIG. 6 is a block diagram of the vehicle-mounted device. FIG. 7 is a diagram illustrating a specific example of the vehicle-mounted speech dictionary. FIG. 8 is a flowchart illustrating a processing procedure executed by the dictionary generation device. FIG. 9A is a flowchart (part 1) illustrating a processing procedure executed by the vehicle-mounted device. FIG. 9B is a flowchart (part 2) illustrating a processing procedure executed by the vehicle-mounted device. FIG. 10 is a diagram illustrating an example of mounting a microphone.

  Hereinafter, a dictionary generation device, a speech recognition system, and a dictionary generation method according to an embodiment will be described in detail with reference to the accompanying drawings. The present invention is not limited by the embodiment.

  First, a dictionary generation method according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an outline of a dictionary generation method. The dictionary generation method is executed by the dictionary generation device 1 shown in FIG.

  The dictionary generation device 1 shown in FIG. 1 generates a speech dictionary used when performing speech recognition. In such a speech dictionary, a phoneme sequence and an acoustic model are associated with each word.

  Incidentally, in the immigration nations such as Europe and the United States, even in the same language (for example, English), pronunciation may be different depending on the speaker. Specifically, a native speaker and a non-native speaker may have different pronunciations even for the same word.

  Further, since a general speech dictionary is generated for native speakers, the recognition accuracy of speech recognition for non-native speakers decreases.

  Therefore, in the dictionary generation method according to the embodiment, the recognition accuracy of voice recognition is improved by adding a phoneme sequence for a non-native speaker to the voice dictionary.

  Specifically, as shown in FIG. 1, first, the dictionary generation device 1 receives an uttered voice of a word by a non-native speaker different from the standard pronunciation (step S1), and generates a phoneme sequence based on the received uttered voice. Generate (Step S2).

  For example, an administrator of the dictionary generation device 1 (hereinafter simply referred to as “administrator”) has a non-native speaker speak a specified word, or speaks a non-native speaker on the Internet. By collecting, the uttered voice of the non-native speaker can be collected.

  Then, when the administrator inputs the voice data of the collected speech voice to the dictionary generation device 1, the dictionary generation device 1 accepts the speech voice.

  A phoneme string is a character string in which phonemes are represented as characters. That is, the dictionary generation device 1 performs a process of transcribing the uttered voice into a phoneme sequence. This makes it easier for the administrator to obtain a phoneme sequence than when a phoneme sequence is transcribed from an uttered voice.

  Thereafter, the dictionary generation device 1 adds the phoneme string to the speech dictionary 41 (Step S3). As a result, the phonetic sequence of the non-native speaker is associated with the phoneme sequence of the native speaker in the voice dictionary 41.

  The example shown in FIG. 1 shows a case where a phoneme string of “Jason” is previously registered as a default (for native) in “Jason” of the voice dictionary 41 in kana notation.

  Then, a case where phoneme strings such as "Jason", "Yason", and "Hathon" are added to "Jason" in kana notation. Therefore, by using the voice dictionary 41, it is possible to recognize the uttered voices of “Jason”, “Yason”, and “Hasson” as “Jason”.

  As described above, in the dictionary generation method according to the embodiment, the speech dictionary 41 is updated by adding the phoneme sequence based on the non-native uttered speech to the speech dictionary 41. As a result, the recognition rate of voice recognition can be improved.

  Further, in the dictionary generation method according to the embodiment, since the phoneme string is generated on the dictionary generation device 1 side, the speech dictionary 41 can be easily updated.

  By the way, as described above, the phonetic dictionary 41 registers a plurality of phoneme strings for one word. Therefore, for example, an approximate phoneme sequence may be registered for a plurality of words. For example, when the words "Jason" and "Yason" (Yason) are registered in the voice dictionary 41, a phoneme string of "Yason" in kana notation is registered in both "Jason" and "Yason". There are cases.

  In this case, since the words “Jason” and “Yason” are extracted from the speech dictionary 41 for the uttered speech “Yason”, the recognition rate of the speech recognition may be reduced.

  For this reason, it is possible to improve the recognition rate of speech recognition by weighting each phoneme string registered in the speech dictionary 41, but this point will be described later in detail.

  Next, the configuration of the speech recognition system S according to the embodiment will be described with reference to FIG. FIG. 2 is a schematic diagram of the speech recognition system S. As shown in FIG. 2, the speech recognition system S includes the dictionary generation device 1 and a plurality of in-vehicle devices 50-1 to 50-n (n is a natural number of 2 or more). Note that the in-vehicle devices 50-1 to 50-n are examples of a voice recognition device. Hereinafter, the in-vehicle devices 50-1 to 50-n will be simply referred to as the in-vehicle device 50.

  The dictionary generation device 1 and the in-vehicle device 50 are connected via a network N, and can perform data communication with each other. Each in-vehicle device 50 is a voice recognition device equipped with a voice dictionary generated and updated by the dictionary generation device 1, and supports voice operations by a user in the vehicle.

  For example, as described later, the user can call the telephone directory of a user terminal such as a smartphone or determine the destination of the navigation device as a voice operation via the in-vehicle device 50.

  Next, a configuration example of the dictionary generation device 1 will be described with reference to FIG. FIG. 3 is a block diagram of the dictionary generation device 1. As shown in FIG. 3, the dictionary generation device 1 includes a communication unit 2, a control unit 3, and a storage unit 4.

  The communication unit 2 is a communication module that performs data communication with each in-vehicle device 50 via the network N. The communication unit 2 can distribute a voice dictionary to each in-vehicle device 50 in accordance with an instruction from the control unit 3 and can acquire a voice dictionary distribution instruction from each in-vehicle device 50.

  The control unit 3 includes a reception unit 30, a generation unit 31, an adjustment unit 32, an update unit 33, and a distribution unit 34. The control unit 3 includes, for example, a computer having a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), a HDD (Hard Disk Drive), an input / output port, and various circuits. .

  The CPU of the computer functions as, for example, the reception unit 30, the generation unit 31, the adjustment unit 32, the update unit 33, and the distribution unit 34 of the control unit 3 by reading and executing a program stored in the ROM.

  In addition, at least some or all of the receiving unit 30, the generating unit 31, the adjusting unit 32, the updating unit 33, and the distributing unit 34 of the control unit 3 may be configured as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). Hardware.

  The storage unit 4 corresponds to, for example, a RAM or an HDD. The RAM and the HDD have a voice dictionary database 40 and a rule information database 42. Note that the dictionary generation device 1 may acquire the above-described programs and various information via another computer or a portable recording medium connected via a wired or wireless network.

  The voice dictionary database 40 is a database that stores the voice dictionary 41 as a table. FIG. 4 is a diagram showing a specific example of the voice dictionary 41. As shown in FIG. 4, the voice dictionary 41 is information in which “name”, “default”, “addition”, and “accent” are associated with each other.

  “Name” is an example of a word. "Default" indicates a phoneme sequence for a native speaker. The “addition” indicates a phoneme string for a non-native speaker added by the control unit 3.

  “Accent” indicates the accent of the speaker who speaks each phoneme sequence. FIG. 4 illustrates native, Nordic, Germanic, and Spanish as accents. These accents are added by the control unit 3 based on rule information described later.

  Returning to FIG. 3, the rule information database 42 will be described. The rule information database 42 is a database that stores, as a table, rule information that is information on the regularity of phoneme strings based on native utterances and phoneme strings based on non-native utterances.

  FIG. 5 is a diagram illustrating an example of the rule information. As shown in FIG. 5, the regularity information 43 is information in which “regularity”, “application target”, “native”, and “non native” are associated with each other.

  “Regularity” includes “classification” indicating a classification of regularity, and “event” corresponding to the “classification”. “Applicable object” indicates a specific example of a word to which regularity is applied, and “native” and “non native” indicate native and non-native pronunciations, respectively.

  Note that, here, for the sake of simplicity of explanation, “native” and “non native” are each shown in kana notation, but they may be shown by phoneme strings.

  As shown in FIG. 5, for example, in "pronunciation of J", native speakers pronounce Jo, Jae, and Ju, whereas Spanish non-native speakers speak Lee, Yo, Ho, and Fu. Pronounced

  For this reason, the native non-speaker pronounces "Jessica" as "Jessica", while the Spanish non-native speaker pronounces "Jessica".

  Regarding "henry", a native speaker who pronounces "henry" pronounces "enri" while a French non-native speaker pronounces "enri" and a Spanish non-native speaker pronounces "enri". Similarly, for "james", a native speaker pronounces "James", a French non-native speaker pronounces "Jams", and a Spanish non-native speaker pronounces "James". It is difficult to estimate the phoneme sequence from the alphabet of “henry” and “james” because the alphabet notation and the pronunciation by a non-native speaker are different.

  As described above, the regularity information 43 is information indicating the regularity of the phoneme string according to the non-native attribute, and by referring to the regularity information 43, it is possible to identify which non-native speaker the user is. It is possible to It should be noted that the regularity information 43 may be input and updated to the dictionary creation device 1 by the administrator, or the dictionary creation device 1 may be provided with a machine learning function, and the dictionary creation device 1 may be provided with the regularity information 43. 43 can be updated as appropriate.

  Returning to the description of FIG. 3, the receiving unit 30 of the control unit 3 will be described. The receiving unit 30 receives an uttered voice of a word with a pronunciation different from the standard pronunciation. That is, the receiving unit 30 receives the uttered voice of the word of the non-native speaker as a voice waveform.

  As described above, the receiving unit 30 can receive the uttered voice collected by the administrator. At this time, the receiving unit 30 receives the input of the character string of the word from the administrator, and notifies the generating unit 31 of the character string of the word indicated by the uttered voice.

  The receiving unit 30 may also receive the input of the native language type of the non-native speaker. In this case, the control unit 3 can update the rule information database 42 based on information on the native language.

  Next, the generation unit 31 will be described. The generation unit 31 generates a phoneme sequence based on the uttered voice received by the reception unit 30. For example, the generation unit 31 has a table indicating a correlation between a speech waveform and a corresponding phoneme sequence, and can generate a phoneme sequence based on the uttered voice by referring to the table. The phoneme string generated by the generation unit 31 is notified to the adjustment unit 32.

  The adjusting unit 32 adjusts the phoneme string generated by the generating unit 31 and notifies the updating unit 33 of the adjusted phoneme sequence. For example, the adjustment unit 32 converts the phoneme string into a voice, and outputs the phoneme string as a reproduced sound from a speaker (not shown) to the administrator. The administrator can manually adjust the phoneme sequence by comparing the reproduced sound with the actual uttered voice. This makes it possible to bring the phoneme sequence closer to the actual uttered voice.

  The adjusting unit 32 can also automatically perform the above processing. That is, the adjustment unit 32 compares the voice waveform based on the phoneme sequence with the voice waveform of the actual uttered voice, and appropriately corrects the phoneme sequence at a portion where the waveform is different.

  Then, the adjusting unit 32 performs such processing until the voice waveform based on the phoneme sequence substantially matches the voice waveform of the actual uttered voice, thereby making it possible to bring the phoneme sequence closer to the actual uttered voice. . As described above, the adjustment unit 32 can accurately reproduce an actual uttered voice using the phoneme sequence by adjusting the phoneme sequence. Therefore, it is possible to improve the speech recognition rate when the speech dictionary 41 is used.

  The updating unit 33 updates the voice dictionary 41 by adding the phoneme sequence notified from the adjusting unit 32 to the voice dictionary 41. The updating unit 33 updates the voice dictionary 41 by associating a phoneme string for a non-native speaker with a phoneme string for a native speaker.

  As a result, the speech dictionary 41 includes a phoneme sequence for a native speaker and a phoneme sequence for each non-native speaker. Therefore, by using the voice dictionary 41, it is possible to recognize both uttered voices of the native speaker and each non-native speaker. That is, it is possible to improve the recognition rate of voice recognition.

  The distribution unit 34 distributes the voice dictionary 41 to each vehicle-mounted device 50 based on a distribution request from each vehicle-mounted device 50. The distribution request includes text registered in an external device to be controlled by each in-vehicle device 50.

  The delivery unit 34 extracts a phoneme string associated with the text from the speech dictionary database 40 and delivers the phoneme string to each in-vehicle device 50. That is, the distribution unit 34 transmits only the phoneme string corresponding to the text registered in the external device. This makes it possible to reduce the data capacity of the vehicle-mounted voice dictionary 70 (see FIG. 6) held by each vehicle-mounted device 50.

  In this way, the distribution unit 34 can improve the convenience of the voice dictionary 41 by optimizing the voice dictionary 41 to be distributed for each in-vehicle device 50.

  Next, the in-vehicle device 50 will be described with reference to FIG. FIG. 6 is a block diagram of the vehicle-mounted device 50. FIG. 6 shows the user terminal 80 and the navigation device 81 together. The user terminal 80 and the navigation device 81 are examples of an external device.

  The user terminal 80 is a terminal having a hands-free call function, such as a smartphone owned by the user. For example, the user terminal 80 notifies the in-vehicle device 50 of the text information of the name registered in the telephone directory and the operation history to the user terminal 80. Further, the user terminal 80 can also make a hands-free call via the in-vehicle device 50.

  The navigation device 81 is a module that guides the traveling route of the vehicle, and notifies the vehicle-mounted device 50 of text information in which a destination can be registered. The navigation device 81 can also set a destination based on a user's voice operation via the in-vehicle device 50.

  The in-vehicle device 50 includes a control unit 6 and a storage unit 7. The control unit 6 includes an acquisition unit 60, a recognition unit 61, a weighting unit 62, and an execution unit 63. The control unit 6 includes, for example, a computer having a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), a HDD (Hard Disk Drive), an input / output port, and various circuits. .

  The CPU of the computer functions as an acquisition unit 60, a recognition unit 61, a weighting unit 62, and an execution unit 63 of the control unit 6, for example, by reading and executing a program stored in the ROM.

  In addition, at least some or all of the acquisition unit 60, the recognition unit 61, the weighting unit 62, and the execution unit 63 of the control unit 3 are implemented by hardware such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). It can also be configured.

  The storage unit 7 corresponds to, for example, a RAM or an HDD. The RAM and the HDD store the vehicle-mounted voice dictionary 70, the specifying information 71, and the command information 72. Note that the in-vehicle device 50 may acquire the above-described programs and various types of information via another computer or a portable recording medium connected via a wired or wireless network.

  The in-vehicle device 50 is connected to a communication device (not shown), and can perform data communication with the dictionary generation device 1 by connecting to the network N (see FIG. 2) via the communication device.

  The acquisition unit 60 acquires a voice dictionary from the dictionary generation device 1. Specifically, the obtaining unit 60 obtains text information that is information relating to text registered in the user terminal 80 to be controlled or the text registered in the navigation device 81, and sends a distribution request of the audio dictionary corresponding to the text information to the dictionary. Transmit to the generation device 1.

  Here, the text refers to a text of a word that is necessary when the user performs a voice operation, such as a name registered in the telephone directory of the user terminal 80 and a place name registered in the navigation device 81. When the acquisition unit 60 acquires a phoneme string corresponding to each text from the dictionary generation device 1, the acquisition unit 60 adds the phoneme string to the storage unit 7 as the vehicle-mounted speech dictionary 70.

  As described above, the acquiring unit 60 acquires the phoneme string based on the text information and adds the acquired phoneme string to the vehicle-mounted speech dictionary 70, so that the data capacity of the vehicle-mounted speech dictionary 70 can be reduced.

  The acquisition unit 60 can also acquire the operation history of the user terminal 80 from the user terminal 80. The operation history includes an utterance history of the user terminal 80 and a browsing history of the web browser of the user terminal 80. These operation histories are notified to a weighting unit 62 described later.

  The recognizing unit 61 performs voice recognition based on the user's uttered voice input from a microphone (not shown), the vehicle-mounted voice dictionary 70, and the specifying information 71. For example, the recognition unit 61 operates in a first mode for performing voice recognition based on the vehicle-mounted voice dictionary 70 and a second mode for performing voice recognition based on the specifying information 71.

  The first mode is a mode for receiving a voice operation by a user based on the vehicle-mounted voice dictionary 70. The user speaks a keyword according to each operation command, and when the recognition unit 61 recognizes the keyword, the mode shifts from the second mode to the first mode.

  Specifically, when the user calls the telephone directory, when the user speaks “Call (XX)”, it is possible to call the telephone directory corresponding to the word following the call (corresponding to the above XX). It is. The recognition unit 61 outputs the word recognized in the first mode to the execution unit 63.

  In the first mode, when a plurality of words are extracted from one uttered voice in the first mode, the recognizing unit 61 can output each of the plurality of words to the weighting unit 62.

  The phoneme sequence registered in the vehicle-mounted speech dictionary 70 is weighted by a weighting unit 62 described later. This makes it possible to suppress erroneous speech recognition by the recognition unit 61. A specific example of the vehicle-mounted speech dictionary 70 will be described later with reference to FIG.

  On the other hand, the second mode is a mode for specifying the “accent” of the user. That is, the recognizing unit 61 can specify the user's accent from the conversation in the vehicle of the user in the second mode.

  Specifically, the recognition unit 61 specifies the accent of the user by referring to the specifying information 71. The specifying information 71 is information for specifying an accent peculiar to each non-active speaker, and is created by the administrator based on the regularity information 43 described above. The recognition unit 61 outputs information on the accent specified in the second mode to the weighting unit 62.

  The weighting unit 62 performs weighting for each phoneme sequence registered in the vehicle-mounted speech dictionary 70. Specifically, the weighting unit 62 weights each phoneme string registered in the vehicle-mounted speech dictionary 70 based on the information on the accent inputted from the recognition unit 61 when the recognition unit 61 operates in the second mode. I do.

  FIG. 7 is a diagram showing a specific example of the vehicle-mounted speech dictionary 70. As shown in FIG. 7, the in-vehicle voice dictionary 70 is information in which “name”, “phoneme sequence”, “accent”, “weight”, and “valid flag” are associated with each other.

  The weighting unit 62 updates the “weight” of each phoneme string based on the information on “accent” input from the recognition unit 61. The example shown in FIG. 7 shows a case where the weight of Northern Europe is “0.7” and the weight of Germanic is “0.3”.

  In other words, the case where the user is likely to be a Nordic or Germanic non-native speaker is high. Therefore, by setting the valid flag of a Nordic or Germanic phoneme string to "1", the phoneme string is made valid, and by setting the valid flags of other phoneme strings to "0", such phoneme string is made. Disable a column.

  Thus, the above-described recognition unit 61 performs voice recognition from a valid phoneme sequence among the phoneme sequences registered in the vehicle-mounted speech dictionary 70. This makes it possible to suppress erroneous speech recognition.

  In addition, when the recognition unit 61 recognizes a plurality of words for one uttered voice, the weighting unit 62 displays the plurality of words on a display screen (not shown) and performs weighting based on a user's selection operation. Is also possible.

  That is, the weighting unit 62 can easily and accurately weight each phoneme string by performing weighting by a user's selection operation. As described above, the weighting unit 62 performs weighting on each phoneme sequence, so that the optimal voice dictionary for each speaker can be the vehicle-mounted voice dictionary 70. Thereby, erroneous recognition can be suppressed while improving the recognition accuracy of voice recognition.

  Note that the weighting unit 62 can also delete the phoneme string for which the valid flag is off from the vehicle-mounted speech dictionary 70. Further, when the user's “accent” is specified, it is also possible to acquire only the phoneme string corresponding to the accent from the dictionary generation device 1.

  This makes it possible to reduce the data capacity of the vehicle-mounted voice dictionary 70. In addition, by obtaining only the phoneme string corresponding to such accent from the dictionary generation device 1, it is possible to suppress the communication capacity.

  Further, the weighting unit 62 can also perform weighting based on the operation history of the user terminal 80. In this case, the weighting unit 62 estimates the user's native language based on the country code included in the telephone number in the speech history and the country code included in the domain of the web browser. Then, the weighting unit 62 can perform weighting for each phoneme string after estimating the “accent” based on the native language.

  Returning to the description of FIG. 6, the execution unit 63 will be described. The execution unit 63 converts the word recognized by the recognition unit 61 into a command in the first mode, and controls the user terminal 80 and the navigation device 81.

  The command information 72 is information about a command for controlling the user terminal 80 and the navigation device 81. The execution unit 63 can convert a word into a command by referring to the command information 72.

  Then, the execution unit 63 outputs the command to the user terminal 80 or the navigation device 81. Accordingly, the user terminal 80 and the navigation device 81 execute an operation based on the command. In other words, the user terminal 80 and the navigation device 81 can receive a voice operation by the user.

  Next, a processing procedure executed by the dictionary generation device 1 will be described with reference to FIG. FIG. 8 is a flowchart illustrating a processing procedure executed by the dictionary generation device 1.

  As shown in FIG. 8, the dictionary generation device 1 first determines whether or not the utterance sound of a word by a non-native speaker has been received (step S101), and if the utterance sound has been received (step S101, Yes), A phoneme sequence for the uttered voice is generated (step S102).

  Subsequently, the dictionary generation device 1 adds the phoneme string to the voice dictionary 41 (step S103), and ends the processing. If the utterance sound has not been received in step S101 (No in step S101), the dictionary generation device 1 ends the process as it is.

  Next, a processing procedure executed by the vehicle-mounted device 50 will be described with reference to FIGS. 9A and 9B. 9A and 9B are flowcharts illustrating a processing procedure executed by the vehicle-mounted device 50. 9A shows a processing procedure of the vehicle-mounted device 50 in the first mode, and FIG. 9B shows a processing procedure of the vehicle-mounted device 50 in the second mode.

  As shown in FIG. 9A, the in-vehicle device 50 first determines whether or not a voice input has been received (step S201). It is determined whether there is one recognized word candidate (step S202).

  Subsequently, when the number of candidates is not one (Step S202, No), the in-vehicle device 50 presents two or more word candidates to the user (Step S203). Subsequently, the in-vehicle device 50 receives a selection operation for a plurality of candidates from the user (step S204), and weights the phoneme string based on the selection operation (step S205).

  Then, the vehicle-mounted device 50 updates the vehicle-mounted voice dictionary (step S206), outputs a command based on the selection operation (step S207), and ends the process. If the in-vehicle device 50 determines that there is only one word candidate in the process of step S202 (step S202, Yes), the process proceeds to step S207. When the voice input is not received (step S201, No), the in-vehicle device 50 ends the process.

  Next, a processing procedure of the vehicle-mounted device 50 in the second mode will be described with reference to FIG. 9B. As shown in FIG. 9B, the in-vehicle device 50 determines whether a voice input has been received (Step S211), and if a voice input has been received (Step S211, Yes), specifies the accent based on the specifying information 71. It is determined whether or not the operation has been completed (step S212).

  When the accent is specified (step S212, Yes), the in-vehicle device 50 weights the phoneme sequence based on the accent (step S213), updates the in-vehicle speech dictionary (step S214), and performs processing. To end.

  If the in-vehicle device 50 has not received a voice input (No at Step S211) or has failed to identify an accent (No at Step S212), the process ends.

  As described above, the dictionary generation device 1 according to the embodiment includes the reception unit 30, the generation unit 31, and the update unit 33. The receiving unit 30 receives an uttered voice of a word with a pronunciation different from the standard pronunciation. The generation unit 31 generates a phoneme sequence based on the uttered voice received by the reception unit 30.

  The updating unit 33 updates the voice dictionary 41 by adding the phoneme sequence generated by the generating unit 31 to the voice dictionary 41. Therefore, according to the dictionary generation device 1 according to the embodiment, the recognition rate of voice recognition can be improved.

  By the way, for example, in a car, a plurality of users may board. For this reason, the in-vehicle device 50 can also perform the weighting for each user by specifying the boarding position for each user.

  FIG. 10 is a diagram illustrating an example of mounting a microphone. As shown in FIG. 10, a plurality of microphones M1 to M4 are mounted on the vehicle C, and each of the microphones M1 to M4 is connected to the vehicle-mounted device 50.

  The microphone M1 is provided in front of the driver's seat and detects a driver's voice. The microphone M2 is provided in front of the passenger seat and detects a voice of a user who gets on the passenger seat. The microphones M3 and M4 are provided on the side wall of the rear seat, respectively, and detect the voice of the user who gets on the rear seat.

  The in-vehicle device 50 can identify the speaker based on the detection result in each of the microphones M1 to M4. That is, when the driver speaks, the voice input from the microphone M1 is relatively louder than the voices input from the other microphones M2 to M4.

  Therefore, in this case, the vehicle-mounted device 50 can recognize that the speaker is the driver. In this case, the in-vehicle device 50 can specify an accent for each speaker and weight the phoneme strings in the in-vehicle speech dictionary 70.

  That is, in such a case, the vehicle-mounted voice dictionary 70 weights each of the microphones M1 to M4. Then, when each user performs a voice input, the in-vehicle device 50 can improve the voice recognition rate by using the in-vehicle voice dictionary 70 tuned to each of the microphones M1 to M4. .

  Here, the case where the in-vehicle device 50 specifies the speaker based on the volume of each of the microphones M1 to M4 has been described, but the present invention is not limited to this. For example, the speaker may be identified by analyzing the mouth of each user in an image captured by a camera that captures the inside of the vehicle.

  Further, in the above-described embodiment, the in-vehicle device 50 has been described as an example of the voice recognition device. However, the present invention is not limited to this. That is, the voice recognition device can be applied to various devices that perform voice recognition, such as a smartphone, a tablet terminal, a personal computer, and a household electrical appliance.

  In the above-described embodiment, the case where the voice dictionary 41 is used for voice recognition has been described, but the present invention is not limited to this. That is, the phoneme sequence registered in the voice dictionary 41 can be applied to text-to-speech. In this case, the text can be read aloud according to the user with an easily recognizable accent.

  Further effects and modifications can be easily derived by those skilled in the art. Thus, the broader aspects of the present invention are not limited to the specific details and representative embodiments shown and described above. Accordingly, various modifications may be made without departing from the spirit or scope of the general inventive concept as defined by the appended claims and equivalents thereof.

Reference Signs List 1 dictionary generation device 30 reception unit 31 generation unit 32 adjustment unit 33 update unit 34 distribution unit 41 voice dictionary 43 regularity information 50 in-vehicle device 60 acquisition unit 61 recognition unit 62 weighting unit 63 execution unit 70 in-vehicle voice dictionary 71 identification information 72 Command information S Voice recognition system

Claims (7)

A receiving unit for receiving a speech sound of a word with a pronunciation different from the standard pronunciation,
A generation unit that generates a phoneme sequence based on the uttered voice received by the reception unit,
An updating unit that updates the voice dictionary by adding the phoneme string generated by the generating unit to the voice dictionary. A speech recognition system comprising: the dictionary generation device according to claim 1; and a speech recognition device that performs speech recognition by the dictionary generation device based on the speech dictionary. The voice recognition device,
The speech recognition system according to claim 2, further comprising: an acquisition unit configured to acquire the phoneme string corresponding to a word registered in an external device. The voice recognition device,
The voice according to claim 2, further comprising a weighting unit configured to weight each of the phoneme strings registered in the speech dictionary based on rule information indicating regularity of the phoneme strings for each of the different pronunciations. Recognition system. The voice recognition device,
A recognition unit configured to recognize a word registered in the voice dictionary based on a voice input;
The weighting unit,
The method according to claim 4, wherein when a plurality of words are extracted by the recognition unit, the plurality of words are presented to a user, and weighting is performed based on a selection operation performed on the plurality of words by the user. Voice recognition system. The weighting unit,
The speech recognition system according to claim 4, wherein the regularity to be applied to the user is selected based on a conversation voice of the user, and the weight of the selected regularity is increased. A receiving step of receiving an uttered voice of a word with a pronunciation different from the standard pronunciation,
A generation step of generating a phoneme sequence based on the uttered voice received by the reception step,
Updating the voice dictionary by adding the phoneme string generated in the generating step to the voice dictionary.

Images