JP5173895B2 - Voice recognition device - Google Patents

Description

  The present invention relates to a speech recognition apparatus.

  Voice recognition devices are used in various fields such as destination search for car navigation. In a speech recognition apparatus, generally, a pattern of each word recorded in a speech recognition dictionary and a speech (word) pattern uttered by a user are matched, and a word having a high degree of coincidence is recognized as a word uttered by the user. . Such voice recognition may cause misrecognition, but if the guidance to the user is uniform (for example, “Please speak again”) for each misrecognition, the same exchange with the user is repeated. Become. Therefore, if the cause of misrecognition is not specified, misrecognition is repeated forever. In view of this, the apparatus described in Patent Document 1 identifies the factors of misrecognition (speech power, speech speed, acoustic characteristics, ambient noise) and presents the factors to the user. Furthermore, when a factor is presented and misrecognition occurs, and the cause of the current misrecognition is the same as the factor of the previous misrecognition, the second factor is presented to the user.

JP 2004-325635 A JP-A-10-133849 JP 2007-264126 A

  As a cause of misrecognition, there are cases where the user uses a word other than the words recorded in the speech recognition dictionary in addition to the above. In this case, a factor is presented to the user as in the above device, and even if the user utters again using the same word while paying attention to the factor, the device correctly recognizes the voice (word) uttered. It is not possible to repeat the recognition error.

  Then, this invention makes it a subject to provide the speech recognition apparatus which suppresses repeating misrecognition.

A speech recognition device according to the present invention is a speech recognition device that recognizes speech uttered by a user based on words recorded in a speech recognition dictionary, and when erroneous recognition occurs , When the same misrecognition occurs again after prompting the recurrence voice while paying attention to the factors, it is determined that the user has uttered a word not recorded in the speech recognition dictionary, and the user is prompted to paraphrase. The cause of the misrecognition to be noted when prompting the recurrent voice is at least one of loudness, speaking timing, and speaking speed.

  In this speech recognition apparatus, words are recorded in the speech recognition dictionary, and speech (words) uttered by the user is recognized based on the recorded word data. When erroneous recognition occurs, the voice recognition device determines that the user has uttered a word that is not recorded in the voice recognition dictionary. As long as the user utters a word that is not recorded in the speech recognition dictionary, the speech recognition device cannot recognize the word correctly, so the user needs to utter using a different word. Therefore, in the speech recognition apparatus, in order to make the user use the words recorded in the speech recognition dictionary, the user is encouraged to paraphrase the words into different words. If the reworded word is recorded in the speech recognition dictionary, the speech recognition apparatus can correctly recognize the word. As described above, in the speech recognition apparatus, when erroneous recognition occurs, it is possible to suppress repeated erroneous recognition by prompting the user to paraphrase. As a result, the reliability of the user's voice recognition apparatus can be improved.

  The voice recognition device of the present invention may be configured to present the second candidate word to the user when erroneous recognition occurs.

  In this speech recognition apparatus, the degree of coincidence between each word recorded in the speech recognition dictionary and the speech (word) uttered by the user is sequentially obtained. First, the first candidate word (the highest degree of coincidence) is given to the user. Word). When a misrecognition occurs in the first candidate word, the speech recognition apparatus presents the second candidate word (word with the next highest matching score) to the user. As a result, if the second candidate word is a word uttered by the user, the speech recognition apparatus can recognize the word correctly. As described above, in the speech recognition apparatus, when erroneous recognition occurs, the second candidate word is also presented to the user, whereby repeated erroneous recognition can be further suppressed.

  The present invention can suppress repeated misrecognition by prompting the user to paraphrase when misrecognition occurs.

It is a block diagram of the speech recognition apparatus which concerns on this Embodiment. It is a flowchart which shows an example of interaction between the speech recognition apparatus which concerns on this Embodiment, and a user. It is a flowchart which shows the other example of interaction between the speech recognition apparatus which concerns on this Embodiment, and a user.

  Hereinafter, embodiments of a speech recognition apparatus according to the present invention will be described with reference to the drawings.

  In this embodiment, the speech recognition apparatus according to the present invention is applied to a speech recognition apparatus mounted on a vehicle. The voice recognition device according to the present embodiment is used for various setting operations in a navigation device, an air conditioner device, an audio device and the like mounted on a vehicle, and a voice (word) uttered by a user (driver or the like) during the various setting operations. ) And outputs the voice (word) that has been correctly recognized to each device.

  A speech recognition apparatus 1 according to the present embodiment will be described with reference to FIG. FIG. 1 is a configuration diagram of a speech recognition apparatus according to the present embodiment.

  The speech recognition apparatus 1 recognizes speech (words) uttered by the user based on word data recorded in the speech recognition dictionary 31, and determines whether or not the recognition result is a misrecognition from a dialogue with the user. . In particular, in the voice recognition device 1, when it is determined that the recognition is erroneous, the guidance for the user is sequentially changed in order to prevent the erroneous recognition from being repeated. For this purpose, the speech recognition apparatus 1 includes a microphone 10, a speaker 20, and an ECU [Electronic Control Unit] 30.

  The microphone 10 is attached to a vehicle interior (particularly, in the vicinity of the front seat) and collects sound (particularly, voice uttered by the user) including air vibrations. When a sound is input, the microphone 10 converts the sound into an electric signal, and transmits the electric signal to the ECU 30 as an input sound signal.

  The speaker 20 is a vehicle-mounted speaker shared with other systems. When the speaker 20 receives the guidance signal from the ECU 30, the speaker 20 outputs a sound according to the guidance signal.

  The ECU 30 is an electronic control unit including a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like, and comprehensively controls the speech recognition apparatus 1. The ECU 30 receives an input audio signal from the microphone 10. When the input voice signal is received during the setting operation in each device of the vehicle, the ECU 30 refers to the voice recognition dictionary 31 and recognizes the voice (word) uttered by the user by the voice recognition engine 32. During the dialogue with the user, the ECU 30 sets the content of the guidance in the dialogue engine 33 based on the recognition result in the voice recognition engine 32 and transmits a guidance signal to the speaker 20. In particular, when it is determined that the recognition is erroneous, the ECU 30 sets the content of the guidance in consideration of points to be noted at the time of re-speech input to the user by the dialog engine 33. In addition, when it is determined that the recognition is correct, the ECU 30 transmits a recognition information signal including information of the word of the correct recognition to the apparatus that is performing the setting operation.

  The voice recognition dictionary 31 is provided in a predetermined area of the storage device of the ECU 30. The voice recognition dictionary 31 records a large number of words that may be used in various settings in each device of the vehicle, and records pattern data (for example, frequency characteristic patterns) for each word. .

  Each time the voice recognition engine 32 receives an input voice signal during various setting operations in each device of the vehicle, the voice recognition engine 32 performs frequency analysis on the voice (electrical signal) input to the microphone 10 and uses the input voice as a frequency. Convert to a characteristic pattern. Then, the speech recognition engine 32 performs matching between the frequency characteristic pattern of the input speech and the recorded word pattern for each word recorded in the speech recognition dictionary 31, and calculates the degree of coincidence. Further, the speech recognition engine 32 determines whether or not the degree of matching of each word is greater than or equal to a threshold value. If there is no word having a degree of matching greater than or equal to the threshold, it is determined as a recognition error. In this case, word candidates recognized in descending order of coincidence are used.

  In the dialog engine 33, when the speech recognition engine 32 determines that the recognition error has occurred, the content of the guidance for prompting the user to input the voice again (for example, “Please speak clearly again”) is set, and the guidance is voiced. A guidance signal for output is transmitted to the speaker 20.

  In the dialogue engine 33, when there is a word whose matching degree is equal to or greater than a threshold, first, the content of the guidance for presenting the recognition result of the first candidate word is set, and a guidance signal for outputting the guidance as a voice is set. It transmits to the speaker 20. When the speech recognition engine 32 recognizes a positive word (for example, “Yes”) as a response from the user after presenting the recognition result of the first candidate word, the dialog engine 33 causes the first candidate It is determined that the word recognition is correct. At this time, the ECU 30 transmits information of the first candidate word as a recognition information signal to the device during the setting operation.

  When the speech recognition engine 32 recognizes a negative word (for example, “No”) as a response from the user after presenting the recognition result with the candidate word, the dialog engine 33 determines that the recognition result is misrecognition. To do. At this time, since the cause of misrecognition is unknown, it is presented to the user in order from what is considered to be a cause of misrecognition. The causes of misrecognition are: loud voice, low voice, fast speaking timing, fast speaking speed, input of words not recorded in the speech recognition dictionary 31, user-specific characteristics (which are difficult to recognize in the first place, ending It is conceivable that the microphone 10 is in an unusable state (timeout or the like). Accordingly, the dialog engine 33 sets the contents of the guidance for re-speech input while paying attention to the factors of misrecognition, and transmits a guidance signal for outputting the guidance to the speaker 20. For example, “Please speak again in a loud voice”, “Please speak slowly again”, and “Please speak in different ways (or words)”. At this time, the previous guidance content is stored, and the guidance with the same content is not presented again. It should be noted that the guidance presentation order in consideration of the cause of misrecognition may be determined in advance, or may be determined depending on the user and the situation at that time.

  If the speech recognition engine 32 recognizes the same word as the previous recognition result again as a response from the user after presenting guidance for re-speech input while paying attention to each cause of misrecognition, the dialog engine 33 Judge that the recognition factor is not loudness, timing or speed of speaking. At this time, if there is a second candidate word as the recognition result, the dialogue engine 33 sets the content of the guidance for presenting the recognition result with the second candidate word, and outputs the guidance as a voice. A guidance signal is transmitted to the speaker 20. When the speech recognition engine 32 recognizes a positive word as a response from the user after presenting the recognition result with the second candidate word, the dialog engine 33 recognizes that the recognition with the second candidate word is correct. Judge. At this time, the ECU 30 transmits information of the second candidate word as a recognition information signal to the device during the setting operation. If the recognition result of the second candidate word is also denied, the same guidance is performed using the third and subsequent candidates when there are third and subsequent candidates.

  On the other hand, if it is determined that the cause of misrecognition is not the loudness of the voice, the timing or speed of speaking, and the candidate word disappears as a recognition result, the dialogue engine 33 records it in the speech recognition dictionary 31 as the cause of misrecognition. It is determined that the user is using a word that has not been made. Then, the dialog engine 33 sets the content of the guidance for changing the way of speaking (or words) and inputting the voice again, and transmits a guidance signal for outputting the guidance to the speaker 20.

  When the speech recognition engine 32 recognizes a candidate word different from the previous recognition result as a response from the user after presenting the guidance for re-speech input in a different way, the dialog engine 33 uses the candidate different from the previous one. The content of the guidance for presenting the recognition result in the word is set, and a guidance signal for outputting the guidance as a voice is transmitted to the speaker 20. When the speech recognition engine 32 recognizes a positive word as a response from the user after presenting a recognition result with a candidate word different from the previous one, the dialog engine 33 determines that the recognition with the candidate word is correct recognition. To do. At this time, the ECU 30 transmits information of the candidate word as a recognition information signal to the device during the setting operation. In addition, after presenting the recognition result, when a negative word is recognized as a response from the user by the voice recognition engine 32, guidance for changing the wording and inputting the voice again is performed.

  It should be noted that each situation used in the dialog engine 33 (for example, when a recognition error is determined, when a recognition result for each candidate word is presented, when loudness is determined as a misrecognition, loudness, speaking timing, speaking speed) The basic text of the guidance corresponding to the case of prompting the user to change the language or the case of prompting the user to change the wording or the word when judging the misrecognition is prepared in advance and stored in a predetermined area of the storage device of the ECU 30 Yes.

  With reference to FIG. 1, the operation in the speech recognition apparatus 1 that communicates with a user who is performing a setting operation by speech recognition will be described. Here, a case where the user is performing destination setting (particularly, destination search) in the navigation device will be described as an example. Two cases will be described. The first case is the case where there is only the first candidate as a recognition result (when the word “cafe” is not recorded in the speech recognition dictionary 31 as a cause of misrecognition). The first case is when there is a second candidate as a recognition result (when the word “cafe” is recognized as the second candidate as a cause of misrecognition). The first case will be described with reference to the flowchart of FIG. 2, and the second case will be described with reference to the flowchart of FIG.

  The first case will be described. The user speaks “Roppongi Cafe” in order to search for a destination (S10). When this sound is collected, the microphone 10 converts the sound into an electric signal and transmits it to the ECU 30 as an input sound signal. When the input speech signal is received, the speech recognition engine 32 of the ECU 30 performs matching between the pattern data of each word recorded in the speech recognition dictionary 31 and the pattern of the input speech, respectively, and sets “1” as the first candidate word. "Roppongi" and "parfait" are recognized (R10). Then, the dialogue engine 33 uses the first candidate words “Roppongi” and “Parfait” to set the guidance content “Do you want to search for Roppongi Parfait?” A signal is transmitted to the speaker 20 (G10). When this guidance signal is received, the speaker 20 outputs a sound “Would you like to search for a Roppongi parfait?” According to the guidance signal (G10).

  Upon hearing this guidance “Would you like to search at Roppongi Parfait?”, The user speaks “No” to deny it (S11). When this sound is collected, the microphone 10 transmits an input sound signal to the ECU 30 as described above. When this input voice signal is received, the voice recognition engine 32 performs pattern matching in the same manner as described above to recognize “No” (R11). Then, the dialog engine 33 determines that “Roppongi” and “parfait” recognized in R10 are misrecognized based on the negative word “No” (J11). At this time, since the cause of the misrecognition is unknown, the dialog engine 33 causes the user to input the voice again while paying attention to the cause of the misrecognition. The guidance content “Please” is set, and a guidance signal for outputting the guidance is transmitted to the speaker 20 (G11). When this guidance signal is received, the speaker 20 outputs a voice “Please change the volume of the voice” in accordance with the guidance signal (G11).

  Upon listening to the guidance “Please change the loudness of the voice”, the user again speaks loudly as “Roppongi Cafe” (S12). When this sound is collected, the microphone 10 transmits an input sound signal to the ECU 30 as described above. When this input speech signal is received, the speech recognition engine 32 performs pattern matching in the same manner as described above, and again recognizes “Roppongi” and “parfait” as the first candidate words (R12). In the dialogue engine 33, since the current recognition result (“Roppongi” and “parfait”) is the same as the previous recognition result (“Roppongi” and “parfait”) in R10, it is determined that the recognition error is erroneous. It is presumed that the recognition factor is the use of a word that is not recorded in the speech recognition dictionary 31 (not the loudness of the voice, the timing of speaking, or the speed) (J12). Therefore, in the dialog engine 33, in order to make the user change the way of speaking and input the voice again, the content of the guidance “Please change the way of speaking. You can search for purposes like ramen in Shibuya.” A guidance signal for outputting the guidance as a voice is transmitted to the speaker 20 (G12). When this guidance signal is received, the speaker 20 outputs a voice saying “Please change the way you speak. You can search for the purpose of ramen like in Shibuya” (G12). .

  The user wanted to drink coffee after listening to this guidance, “Please speak in different ways. You can search for purposes like ramen in Shibuya.” “I want to drink coffee in Roppongi.” (S13). When this sound is collected, the microphone 10 transmits an input sound signal to the ECU 30 as described above. When this input speech signal is received, the speech recognition engine 32 performs pattern matching in the same manner as described above, and recognizes “Roppongi” and “coffee” as the first candidate words (R13). Then, the dialogue engine 33 uses the first candidate words “Roppongi” and “Coffee” to set the guidance content “Do you want to search for Roppongi Coffee?”, And the guidance is for voice output. A signal is transmitted to the speaker 20 (G13). When this guidance signal is received, the speaker 20 outputs a sound “Would you like to search for Roppongi coffee?” In response to this guidance signal (G13).

  Upon listening to the guidance “Do you want to search for coffee in Roppongi?”, The user speaks “Yes” to affirm it (S14). When this sound is collected, the microphone 10 transmits an input sound signal to the ECU 30 as described above. When this input voice signal is received, the voice recognition engine 32 performs pattern matching in the same manner as described above to recognize “Yes” (R14). Then, the dialog engine 33 determines that “Roppongi” and “coffee” recognized in R13 are correct recognition based on the positive word “Yes” (J14). Then, the ECU 30 transmits a recognition information signal including “Roppongi” and “coffee” as a voice recognition result to the navigation device.

  The second case will be described. In this second case, the operation up to R22 for recognizing “Roppongi” and “parfait” for the second time by the speech recognition engine 32 is the same as that in the first case. However, the speech recognition engine 32 recognizes “cafe” as the second candidate after the first candidate “parfait” as the recognition result.

  In the dialog engine 33, since the current recognition result (“Roppongi” and “parfait”) is the same as the previous recognition result (“Roppongi” and “parfait”) in R20, it is determined as misrecognition (J22). Here, since there is “cafe” as the second candidate word, the dialog engine 33 excludes the first candidate “parfait” from the candidates, and sets “Roppongi” and the second candidate word “cafe”. The guidance content “Do you want to search in Roppongi Cafe?” Is set and a guidance signal for outputting the guidance as a voice is transmitted to the speaker 20 (G22). When this guidance signal is received, the speaker 20 outputs a sound “Would you like to search at Roppongi Cafe?” According to this guidance signal (G22).

  Upon hearing the guidance “Do you want to look for at Roppongi Cafe?”, The user utters “Yes” to affirm it (S23). When this sound is collected, the microphone 10 transmits an input sound signal to the ECU 30 as described above. When this input voice signal is received, the voice recognition engine 32 performs pattern matching in the same manner as described above to recognize “Yes” (R23). Based on the positive word “yes”, the dialog engine 33 determines that “Roppongi” recognized in R23 and the second candidate “cafe” are correct recognition (J23). Then, the ECU 30 transmits a recognition information signal including “Roppongi” and “cafe” as a voice recognition result to the navigation device.

  According to this voice recognition device 1, it is determined that the recognition is erroneous, and if the cause of the erroneous recognition cannot be specified, the content of the guidance is sequentially changed to allow the user to input the voice again in consideration of the cause of the erroneous recognition. It is possible to suppress erroneous recognition. As a result, the reliability of the user's voice recognition apparatus can be improved. In particular, according to the voice recognition device 1, even when a user inputs a voice using a word that is not recorded in the voice recognition dictionary 31 by performing guidance that prompts the user to paraphrase, a word different from the previous one is given to the user. Can be used for voice input again. Moreover, according to the speech recognition apparatus 1, when there are also recognition results of the second and subsequent candidates, the second and subsequent candidate words are also presented to the user in order to further suppress repeated erroneous recognition. it can.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is implemented in various forms, without being limited to the said embodiment.

  For example, in the present embodiment, the present invention is applied to a voice recognition device mounted on a vehicle, but can be applied to various other fields.

  In addition, in the present embodiment, an example of the content and order of guidance when it is determined to be misrecognized is shown, but the content and order of guidance are not particularly limited, and content for preventing repeated misrecognition. Or any order. For example, when it is determined that the recognition is wrong, it may be urged not to change the volume of the voice, the timing or speed of speaking, but to change the way of speaking or the word from the beginning.

  Also, in this embodiment, when there are a plurality of word candidates as recognition results, the first candidate word is determined to be misrecognized, and the second and subsequent candidate words are presented to the user. Even when there are a plurality of candidates, the second and subsequent candidate words may not be presented to the user.

  DESCRIPTION OF SYMBOLS 1 ... Voice recognition apparatus, 10 ... Microphone, 20 ... Speaker, 30 ... ECU, 31 ... Voice recognition dictionary, 32 ... Voice recognition engine, 33 ... Dialog engine

Claims (3)

A speech recognition device for recognizing speech uttered by a user based on words recorded in a speech recognition dictionary,
When misrecognition occurs, if the same misrecognition occurs again after prompting the user to pay attention to the cause of misrecognition and then recite, it is determined that the user has uttered a word not recorded in the speech recognition dictionary A speech recognition apparatus that prompts the user to paraphrase.   The speech recognition apparatus according to claim 1, wherein when a recognition error occurs, the second candidate word is presented to the user. The voice according to claim 1 or 2, wherein a factor of misrecognition to be noted when prompting the recurrent voice is at least one of loudness, speaking timing, and speaking speed. Recognition device.

Images