JP6824554B2 - Speech recognition system - Google Patents

Description

The present invention relates to a voice recognition technique for recognizing a user's spoken voice.

As a voice recognition technology for recognizing a user's voice, a first voice recognition that recognizes a user's voice picked up by a microphone in a system equipped with an audio device that outputs the voice of audio content such as music from a speaker. In addition to the unit, a second voice recognition unit that performs voice recognition of the voice output by the audio device to the speaker is provided, and the result of voice recognition by the first voice recognition unit and the result of voice recognition by the second voice recognition unit. There is known a technique for invalidating the result of voice recognition by the first voice recognition unit when the above is the same (for example, Patent Document 1).

According to such a technique, the result of the voice recognition of the first voice recognition unit, which is also performed for the output voice of the audio device that wraps around the microphone, is misidentified as the result of the voice recognition of the user's spoken voice. It can be prevented from doing so.

In addition, as a voice recognition technology for recognizing a user's voice, in a system that performs voice recognition of commands, the first method is to change a set of commands that are candidates for voice recognition according to the command that was previously voice-recognized. The voice recognition means performs voice recognition of the commands in the set of commands that are candidates for voice recognition, while the second voice recognition means performs voice recognition of the commands in the set of commands that were candidates for voice recognition last time. By doing so, in addition to voice recognition of the commands of the set of commands that are candidates for voice recognition this time, there is also a technology that also performs voice recognition of the commands in the set of commands that were candidates for voice recognition last time that the user rephrased. It is known (for example, Patent Document 2).

Jitsuto 2602342 International Publication No. 2011/016129

By the way, when voice recognition is performed for a plurality of recognition candidates that are words such as words and sentences, erroneous recognition is likely to occur for recognition candidates having similar pronunciations.
Therefore, an object of the present invention is to enable more appropriate voice recognition even when there are recognition candidates having similar pronunciations when performing voice recognition for a plurality of recognition candidates.

In order to achieve the above object, the present invention includes a microphone, a first voice recognition dictionary in which a plurality of first recognition candidates that are words are registered, and the microphone in a voice recognition system that recognizes a voice spoken by a user. The first recognition candidate that inputs the picked-up voice and detects the first recognition candidate that matches the input voice better than a predetermined level from the plurality of first recognition candidates registered in the first voice recognition dictionary as the target candidate. Voice recognition means, second voice recognition means,
A second voice recognition dictionary used by the second voice recognition means and a recognition means are provided. Here, the voice recognition system has a single or a plurality of recognition modes including a first recognition mode, and the second voice recognition dictionary used by the second voice recognition means in the first recognition mode , The word excerpt part, which is a part whose pronunciation is not common to other words of a plurality of words having a partially common pronunciation, which is registered as a first recognition candidate in the first speech recognition dictionary, is a second recognition candidate. In the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone, and among the plurality of second recognition candidates registered in the second voice recognition dictionary. Therefore, a second recognition candidate that matches the input voice better than a predetermined level is detected as a target candidate, and in the first recognition mode, the recognition means is registered as the first recognition candidate in the first voice recognition dictionary. For each of the words, the first recognition candidate, which is the word, and the second recognition candidate, which is an excerpt of the word, detected as the target candidates by the first voice recognition means and the second voice recognition means. The number of words is calculated, and the word with the largest calculated number is recognized as the word spoken by the user.

According to such a speech recognition system, the evaluation of the consistency with the spoken speech focusing only on the part whose pronunciation is different from that of other words was added to consideration by using the first speech recognition dictionary and the second speech recognition dictionary. By comprehensive evaluation, it is possible to perform speech recognition of words registered in the first speech recognition dictionary in a form of more strongly identifying a plurality of words having similar pronunciations.

Further, in order to achieve the above object, the present invention has a voice recognition system that recognizes a voice spoken by a user, a plurality of microphones, a plurality of first recognition candidates that are words, and a plurality of corresponding first recognition candidates. The first voice recognition dictionary registered together with the first threshold value and the voice picked up by the microphone are input, and the corresponding first recognition candidate is selected from the plurality of first recognition candidates registered in the first voice recognition dictionary. A second voice recognition means used by a first voice recognition means, a second voice recognition means, and a second voice recognition means that detects a first recognition candidate that matches a voice input better than a level indicated by one threshold as a target candidate. It is equipped with a voice recognition dictionary and recognition means. Here, the voice recognition system has a single or a plurality of recognition modes including a first recognition mode, and the second voice recognition dictionary used by the second voice recognition means in the first recognition mode , The same word as the plurality of words registered as the first recognition candidate in the first voice recognition dictionary is registered as the second recognition candidate together with the plurality of second threshold values corresponding to each of the second recognition candidates. In the first recognition mode, the second threshold value corresponding to each second recognition candidate registered in the second voice recognition dictionary is the same word as the word registered as the second recognition candidate. It shows a level of matching better than the first threshold value corresponding to the first recognition candidate, and in the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone. , The second recognition candidate that matches the input voice better than the level indicated by the corresponding second threshold value is detected as the target candidate from the plurality of second recognition candidates registered in the second voice recognition dictionary. In the first recognition mode, the recognition means uses the first voice recognition means and the second voice recognition means for each word registered as a first recognition candidate in the first voice recognition dictionary. The number of the first recognition candidate and the second recognition candidate, which are the words detected as the target candidates, is calculated, and the word with the largest calculated number is recognized as the word spoken by the user.

According to such a speech recognition system, a plurality of words having similar pronunciations can be obtained by comprehensive evaluation in consideration of evaluation by different thresholds using the first speech recognition dictionary and the second speech recognition dictionary. It is possible to perform voice recognition of words registered in the first voice recognition dictionary in a form of more accurate identification.

Further, in order to achieve the above object, the present invention has a voice recognition system that recognizes a voice spoken by a user, a plurality of microphones, a plurality of first recognition candidates that are words, and a plurality of corresponding first recognition candidates. The first voice recognition dictionary registered together with the first threshold value and the voice picked up by the microphone are input, and the corresponding first recognition candidate is selected from the plurality of first recognition candidates registered in the first voice recognition dictionary. The first voice recognition means, the second voice recognition means, and the second voice recognition means used by the first voice recognition means, the second voice recognition means, and the second voice recognition means, which detect the first recognition candidate that matches the input voice better than the level indicated by the threshold value as the target candidate. 2 A voice recognition dictionary and a recognition means are provided. Here, the voice recognition system has a single recognition mode including a first recognition mode, and the second voice recognition dictionary used by the second voice recognition means in the first recognition mode includes the second voice recognition dictionary. A word excerpt portion that is registered as a first recognition candidate in the first voice recognition dictionary and is a part whose pronunciation is not common to other words of a plurality of words having a partially common pronunciation, and the first voice recognition. A plurality of words registered as first recognition candidates in the dictionary and the same words are registered as second recognition candidates together with a plurality of second threshold values corresponding to each of the second recognition candidates. In the first recognition mode, the second threshold value corresponding to the second recognition candidate, which is the same word as the word registered as the first recognition candidate, registered in the second voice recognition dictionary is the second threshold value. 2 The level of matching is better than the first threshold value corresponding to the first recognition candidate, which is the same word as the word registered as the recognition candidate, and in the first recognition mode, the second voice recognition means. Inputs the voice picked up by the microphone and matches the voice input better than the level indicated by the corresponding second threshold value from among the plurality of second recognition candidates registered in the second voice recognition dictionary. The second recognition candidate is detected as a target candidate, and in the first recognition mode, the recognition means recognizes the first voice for each word registered as the first recognition candidate in the first voice recognition dictionary. The number of the first recognition candidate which is the word, the second recognition candidate which is the word, and the second recognition candidate which is an excerpt part of the word detected as the target candidate by the means and the second voice recognition means. The word that is calculated and the calculated number is the largest is recognized as the word spoken by the user.

Here, in such a voice recognition system, in the first recognition mode, when the recognition means has a plurality of words having the maximum calculated number, among the words having the maximum calculated number. The second recognition candidate, which is an excerpt of the word, may be configured to recognize the word detected as the target candidate as the word spoken by the user.

Further, in this case, in the first recognition mode of the voice recognition system, the recognition means has a plurality of words having the maximum calculated number, and among the words having the maximum calculated number. When there are a plurality of words in which the second recognition candidate which is the excerpt part of the word is detected as the target candidate, the calculated number is the largest word and the second recognition candidate which is the excerpt part of the word. Among the words whose candidates are detected as the target candidates, the first threshold value corresponding to the first recognition candidate which is the word detected as the target candidate and the word detected as the target candidate. A word in which the minimum value between the second threshold value corresponding to the second recognition candidate and the second threshold value corresponding to the second recognition candidate, which is an excerpt portion of the word detected as the target candidate, is the minimum. May be configured to be recognized as a word spoken by the user.

According to such a speech recognition system, evaluation by different thresholds using the first speech recognition dictionary and the second speech recognition dictionary, and consistency with spoken speech focusing only on the part whose pronunciation is different from other words. The speech recognition of the words registered in the first speech recognition dictionary can be performed in a form of more accurately identifying a plurality of words having similar pronunciations by the comprehensive evaluation in consideration of the evaluation of.

Further, in order to achieve the above object, the present invention comprises a voice recognition system that recognizes a voice spoken by a user, a microphone, a first voice recognition dictionary in which a plurality of first recognition candidates that are words are registered, and the above. The voice picked up by the microphone is input, and the first recognition candidate that matches the input voice better than a predetermined level is detected as the target candidate from among the plurality of first recognition candidates registered in the first voice recognition dictionary. It includes a first voice recognition means, a second voice recognition means, a second voice recognition dictionary used by the second voice recognition means, and a recognition means. Here, the voice recognition system has a single or a plurality of recognition modes including a first recognition mode, and the second voice recognition dictionary used by the second voice recognition means in the first recognition mode , The word excerpt part, which is a part whose pronunciation is not common to other words of a plurality of words having a partially common pronunciation, which is registered as a first recognition candidate in the first speech recognition dictionary, is a second recognition candidate. In the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone, and among the plurality of second recognition candidates registered in the second voice recognition dictionary. Therefore, a second recognition candidate that matches the input voice better than a predetermined level is detected as a target candidate, and in the first recognition mode, the recognition means is detected as the target candidate by the first voice recognition means. When there are a plurality of words that are the first recognition candidates, the second recognition candidate, which is an excerpt of the word, is detected as the target candidate by the second voice recognition means. Recognize the word as a word spoken by the user.

According to such a speech recognition system, the evaluation of the consistency with the spoken speech focusing only on the part whose pronunciation is different from that of other words was added to consideration by using the first speech recognition dictionary and the second speech recognition dictionary. By comprehensive evaluation, it is possible to perform speech recognition of words registered in the first speech recognition dictionary in a form of more accurately identifying a plurality of words having similar pronunciations.

Further, in the above voice recognition system, the microphone is arranged in a space where the sound output from the audio source device is emitted from the speaker to the speaker, and the voice recognition system has the first recognition mode and the first recognition mode. A plurality of recognition modes including the second recognition mode are provided, and the second voice recognition dictionary used by the second voice recognition means in the second recognition mode is registered in the first voice recognition dictionary as a first recognition candidate. It is assumed that the same words as the plurality of words are registered as the second recognition candidates, and in the second recognition mode, the voice output by the audio source device to the speaker is input in the second voice recognition means. Then, from among the plurality of second recognition candidates registered in the second voice recognition dictionary, the second recognition candidate that matches the input voice better than a predetermined level is detected as the target candidate, and the second recognition mode In the recognition means, the same word as the second recognition candidate detected as the target candidate during a predetermined period after the second recognition candidate is detected as the target candidate by the second voice recognition means. 1 Even if the recognition candidate is detected as the target candidate by the first voice recognition means, the word that is the first recognition candidate detected as the target candidate is suppressed from being recognized as the word spoken by the user. The word that is the first recognition candidate detected as the target candidate by the first voice recognition means may be configured to be recognized as the word spoken by the user.

Such a voice recognition system is a storage means for storing a plurality of voice recognition data representing the first voice recognition dictionary and the second voice recognition dictionary in the voice recognition system, and a voice input receiving means for receiving a user's voice input. The recognition mode control means for controlling the recognition mode is provided, and the first voice recognition dictionary and the second voice recognition dictionary represented by the predetermined voice recognition data are set in the voice input receiving means, and the recognition is performed. The standby state in which the word recognized by the means is accepted as a voice input, and the first voice recognition dictionary represented by the voice recognition data determined according to the word received as the voice input last time in the first voice recognition dictionary and the second voice recognition dictionary. While updating to the second voice recognition dictionary, a voice input sequence executing state is provided in which a sequence of receiving the word recognized by the recognition means as voice input once or a plurality of times is executed, and the voice input receiving means is in the standby state. If the word is accepted as a voice input in, the state transitions to the voice input sequence executing state, and if the execution of the sequence in the voice input sequence execution state is completed, the state transitions to the standby state, and the recognition mode control means. In the above, when the voice input receiving means is in the standby state, the second recognition mode is set as the recognition mode, and when the voice input receiving means is in the voice input sequence executing state, the recognition mode is set. , The first recognition mode and the second recognition mode may be configured to switch according to the voice recognition data determined according to the word received as the previous voice input.

Alternatively, such a voice recognition system is a storage means that stores a plurality of voice recognition data representing the first voice recognition dictionary and the second voice recognition dictionary in the voice recognition system, and a voice input that receives a user's voice input. A reception means and a recognition mode control means for controlling the recognition mode are provided, and the first voice recognition dictionary and the second voice recognition dictionary represented by predetermined voice recognition data are set in the voice input reception means. The first voice recognition represented by a standby state in which a word recognized by the recognition means is accepted as a voice input and voice recognition data determined according to a word received as a voice input last time in the first voice recognition dictionary and the second voice recognition dictionary. While updating to the dictionary and the second voice recognition dictionary, a voice input sequence executing state is provided in which a sequence of receiving the word recognized by the recognition means as voice input once or multiple times is executed. If a word is accepted as a voice input in the standby state, the state transitions to the voice input sequence executing state, and if the execution of the sequence in the voice input sequence execution state is completed, the transition to the standby state is performed and the recognition In the mode control means, when the voice input receiving means is in the standby state, the second recognition mode is set as the recognition mode, and when the voice input receiving means is in the voice input sequence executing state, the second recognition mode is set. The first recognition mode may be set as the recognition mode.

Further, the above voice recognition system may be a voice recognition system used for voice input in an in-vehicle system mounted on an automobile.

As described above, according to the present invention, when performing voice recognition for a plurality of recognition candidates, it is possible to perform voice recognition more appropriately even when there are recognition candidates having similar pronunciations.

It is a block diagram which shows the structure of the information processing system which concerns on embodiment of this invention. It is a figure which shows the recognition data which concerns on embodiment of this invention. It is a flowchart which shows the voice input setting process which concerns on embodiment of this invention. It is a figure which shows the voice recognition method of the voice recognition engine which concerns on embodiment of this invention. It is a flowchart which shows the audio cancel mode recognition process which concerns on embodiment of this invention. It is a flowchart which shows the similar word strong identification mode recognition process which concerns on embodiment of this invention. It is a figure which shows the transition example of the display screen and a voice recognition dictionary which concerns on embodiment of this invention.

Hereinafter, embodiments of the present invention will be described by taking application to an information processing system mounted on an automobile as an example.
FIG. 1 shows the configuration of the information processing system according to the present embodiment.
As shown in the figure, the information processing system includes a data processing unit 1, a dictionary DB 2, a microphone 3, a voice input unit 4, a speaker 5, an audio source 6, a display device 7, and other peripheral devices 8 such as a GPS receiver. There is.

Here, the audio source 6 is a device that operates under the control of the data processing unit 1 and serves as a sound source for a radio receiver, a music player, or the like, and outputs the audio of the audio content to the speaker 5 and the audio input unit 4. Further, the speaker 5 radiates the sound input from the audio source 6 into the vehicle.

Further, the voice input unit 4 recognizes the voice of the user input from the microphone 3 and outputs the recognition result to the data processing unit 1. The data processing unit 1 uses the voice input unit 4 as a voice input such as a command. , Various processes are performed while using the display device 7 for displaying the screen.
Further, the voice input unit 4 includes a first voice recognition engine 41, a first voice recognition dictionary 42, a second voice recognition engine 43, a second voice recognition dictionary 44, and a recognition adjustment unit 45.

Here, such an information processing system may be configured by using a computer provided with a CPU, a memory, a peripheral device, or the like. In this case, the data processing unit 1 and the voice input unit 4 described above may be used. It may be realized by the CPU executing a computer program.
Next, as shown in FIG. 2, the dictionary DB 2 stores recognition data of a plurality of layers from the first layer recognition data to the third layer recognition data.

Then, in the recognition data of each layer, either the audio cancel mode or the similar word strong identification mode is registered as the recognition mode. In addition, a main voice recognition dictionary and a sub voice recognition dictionary are registered in the recognition data of each layer.
The main speech recognition dictionary is a dictionary for speech recognition, and a number (No.), a word, and a threshold value Th are registered for each of a plurality of words.
By the way, the recognition mode of the first layer recognition data is set to the audio cancel mode, and the main voice recognition dictionary of the first layer recognition data is a plurality of candidates for recognition of voice recognition in the voice input unit 4 in the initial state. The word, its number (No.), and the threshold Th are registered.

A plurality of second layer recognition data can be provided corresponding to each word registered in the main speech recognition dictionary of the first layer recognition data, and the third layer recognition data can be provided for each second layer recognition data. A plurality of words can be provided corresponding to each word registered in the main speech recognition dictionary. That is, the dictionary DB2 has a tree structure in which the recognition data of each layer is used as a node.

Further, in the main voice recognition dictionary of the second layer recognition data, the words of the main voice recognition dictionary of the first layer recognition data corresponding to the second layer recognition data are calculated as the recognition result by the voice recognition in the voice input unit 4. Then, a plurality of words as recognition candidates for voice recognition in the voice input unit 4, their numbers (No.), and a threshold value Th are registered.

Further, in the main voice recognition dictionary of the third layer recognition data, the words of the main voice recognition dictionary of the second layer recognition data corresponding to the third layer recognition data are calculated as the recognition result by the voice recognition in the voice input unit 4. Then, a plurality of words as recognition candidates for voice recognition in the voice input unit 4, their numbers (No.), and a threshold value Th are registered.

Next, whether the recognition mode of each of the second layer recognition data and each of the third layer recognition data is the audio cancel mode or the similar word strong identification mode is pronounced in the main speech recognition dictionary of the recognition data. It is determined by whether or not similar similar words are included, and when the main speech recognition dictionary of the recognition data contains multiple words with similar pronunciations, the recognition mode is set to the similar word strong identification mode. Set to, and in other cases set the recognition mode to audio cancel mode.

Then, as the sub-speech recognition dictionary of the recognition data of each layer in which the recognition mode is set to the audio cancel mode, a dictionary obtained by increasing the threshold value Th of each word of the main speech recognition dictionary of the recognition data by a predetermined value is registered. .. That is, in the main speech recognition dictionary and the sub speech recognition dictionary of the recognition data in which the recognition mode is set to the audio cancel mode, the threshold value Th of each word of the sub speech recognition dictionary is a predetermined value larger than that of the main speech recognition dictionary. Only the points are different.

Further, in the sub-speech recognition dictionary of the recognition data in which the recognition mode is set to the similar word strong identification mode, other words registered in the main speech recognition dictionary of the recognition data and registered in the main speech recognition dictionary. The same word as each word whose pronunciation is similar to, or other words of each word whose pronunciation is similar to other words registered in the main speech recognition dictionary of the recognition data. Register the part whose pronunciation is not similar to that as a word.

That is, for example, as shown in the third layer recognition data of FIG. 2, "go to the first", "go to the second", "go to the second", "yotsume" registered in the main speech recognition dictionary. For the five words that have similar pronunciations to "niiku" and "izutsume niiku", these five words and the parts that differ from the other words in each word are "human", "lid", and "mi". , "Yo", and "when" are registered as words in the sub-speech recognition dictionary. Similarly, the two words whose pronunciations are similar to those of the "next page and the previous page" registered in the main speech recognition dictionary are different from the other words of these two words. Register "next" and "mae" in the secondary speech recognition dictionary.

In addition, in addition to the number (No.) and threshold value Th of each word, the reference number (ref) and partial flag (P) are added to the sub-speech recognition dictionary of the recognition data in which the recognition mode is set to the similar word strong identification mode. to register. Here, the threshold value Th of each word of the sub-speech recognition dictionary is set to a value smaller than the threshold value Th of the word of the main speech recognition dictionary that matches the word or is a part of the word. Further, in the reference number (ref) of each word of the sub-speech recognition dictionary, a word number (No.) of the main speech recognition dictionary that matches the word or is a part of the word is set. Further, in the partial flag (P) of each word of the sub-speech recognition dictionary, a value (Y) indicating that the word is a part of the word of the main speech recognition dictionary is set.

The dictionary DB2 has been described above.
In the above, the case where the recognition data of three layers from the first layer recognition data to the third layer recognition data is provided as the recognition data to be registered in the dictionary DB2 has been described, but the recognition data to be registered in the dictionary DB2 is 2. It may be the recognition data of any number of layers as described above.

By the way, the data processing unit 1 is provided with various functions such as a car navigation function and a music player function, and when activated, a predetermined information processing (for example, a guide map for car navigation is displayed on the display device 7 by the car navigation function). Information processing for displaying, information processing for outputting music from the audio source 6 by the music player function, etc.) is started.
In addition, the data processing unit 1 performs the following processing regarding the reception of voice input.
That is, when the data processing unit 1 is activated, the voice input setting process is started, the first layer recognition data is set as the current recognition data, and the standby state waits for the input of the recognition result from the voice input unit 4. Become. Then, if the recognition result is output from the voice input unit 4 in the standby state, the input of the recognition result is accepted and the transition to the sequence execution state is performed, and in the sequence execution state, the input is received according to the recognition result word. While performing the processing and switching of the current recognition data by setting the recognition data in the next lower layer of the current recognition data as the current recognition data, which corresponds to the recognition result word that received the input of the current recognition data. , The sequence of receiving the input of the recognition result from the voice input unit 4 a predetermined number of times is executed. Then, when the sequence is completed, the first layer recognition data is set as the current recognition data, and the standby state is restored.

Further, the data processing unit 1 performs the above-mentioned voice input setting process as follows.
FIG. 3 shows the procedure of this voice input setting process.
As shown in the figure, in the voice input setting process, the data processing unit 1 monitors the occurrence of the setting of the current recognition data performed as described above (step 302).
Then, when the setting of the current recognition data occurs (step 302), it is checked whether or not the recognition mode of the current recognition data is the audio cancel mode (step 304).

If the recognition mode is the audio cancel mode (step 304), the audio cancel mode is set in the recognition adjustment unit 45 (step 306), and the main voice recognition dictionary of the current recognition data is set in the first voice recognition dictionary 42. , The sub-speech recognition dictionary of the current recognition data is set in the second speech recognition dictionary 44 (step 308).

Then, the recognition adjustment unit 45 is instructed to start voice recognition (step 310), and the process returns to the monitoring of step 302.
On the other hand, when the recognition mode of the current recognition data is not the audio cancel mode but the similar word strong identification mode (step 304), the similar word strong identification mode is set in the recognition adjustment unit 45 (step 312), and the current recognition data is used. The main speech recognition dictionary of the recognition data is set in the first speech recognition dictionary 42, and the sub speech recognition dictionary of the current recognition data is set in the second speech recognition dictionary 44 (step 308).

Then, the recognition adjustment unit 45 is instructed to start voice recognition (step 310), and the process returns to the monitoring of step 302.
The voice input setting process performed by the data processing unit 1 has been described above.
Next, the operation of voice recognition performed by the first voice recognition engine 41 and the second voice recognition engine 43 will be described.
The first voice recognition engine 41 and the second voice recognition engine 43 calculate the score of each recognition candidate word stored in the voice recognition dictionary for the recognition target voice in parallel with the input of the recognition target voice.

That is, the first speech recognition engine 41 calculates the score of the word stored in the first speech recognition dictionary 42 for the recognition target speech in parallel with the input of the recognition target speech, and the second speech recognition engine 43 recognizes it. In parallel with the input of the target voice, the score of each word stored in the second voice recognition dictionary 44 for the recognition target voice is calculated.

The recognition target voice of the first voice recognition engine 41 is a voice input from the microphone 3. On the other hand, with respect to the second voice recognition engine 43, one of the voice input from the microphone 3 and the voice input from the audio source 6 can be selectively used as the recognition target voice of the first voice recognition engine 41.

Here, the score of each word registered in the speech recognition dictionary for the recognition target voice represents a predicted value of the magnitude of the difference between the word and the phrase represented by the recognition target voice and the word, and predicts a larger difference. The more you do, the higher your score.

More specifically, in the calculation of the score, after setting a predetermined initial value as the score, each time the sound of each voice section of the recognition target voice (for example, the voice section for each phonetic element) is input, Whether or not the sound of the voice section matches the pronunciation of the part corresponding to the voice section of each word registered in the voice recognition dictionary is calculated, and the score is reduced by a predetermined value for the matching words. For inconsistent words, the score is increased by a predetermined value. The increase / decrease value of the word score for each voice section of the voice to be recognized is, for example, the ratio of the time length of the voice section to the time length of all voice sections multiplied by the initial value of the score. Sato.

According to such voice recognition, when the recognition target voice is "aiueoka", the transition of the score calculated for the word "aiueo" is calculated for the word "aiueo" in FIG. 4a. As shown in FIG. 4b, while the sound of the recognition target voice that matches the word is input, the score with the word gradually decreases and the sound of the recognition target voice that does not match the word is input. While you are there, your ward score will increase over time.

That is, for example, as shown in FIG. 4a, the recognition target voice "aiueoka", the word "aiueo", and the score gradually decrease during the period in which the recognition target voice "aiueo" sound is input, and then gradually decrease. , Increases when "ka" of the recognition target voice is input.

Similarly, as shown in FIG. 4b, the recognition target voice "aiueoka", the word "aiai" and the score gradually decrease during the period in which the recognition target voice "ai" sound is input. After that, the period during which the recognition target voice "Eoka" is input gradually increases.

By the way, in the first voice recognition engine 41 and the second voice recognition engine 43, the score of the recognition target voice calculated as described above and one of the words is registered in the voice recognition dictionary of the word. When the threshold value becomes Th or less, the hit of the word whose score becomes the threshold value Th or less is detected, and the number (No.) of the hit word is notified to the recognition adjustment unit 45 as hit data.

That is, for example, in the case shown in FIG. 4a, the score for the word "aiueo" is equal to or less than the threshold value Th immediately before the "e" of the recognition target voice "aiueoka" is input. , The hit of the word "aiueo" is detected.

On the other hand, in the case shown in FIG. 4b, since the score for the word "Ai Aio" does not fall below the threshold value Th, the hit of this word "Ai Aio" is not detected.
Next, the operation of the recognition adjustment unit 45 of the voice input unit 4 will be described.
If the data processing unit 1 instructs the start of voice recognition by the voice input setting process shown in FIG. 3, the recognition adjustment unit 45 executes the audio cancel mode recognition process when the audio cancel mode is set. When the similar word strong identification mode is set, the similar word strong identification mode recognition process is executed.

First, the audio cancel mode recognition process performed by the recognition adjustment unit 45 when the audio cancel mode is set will be described.
FIG. 5 shows the procedure of the audio cancel mode recognition process shown.
As shown in the figure, in this audio cancel mode recognition process, first, the recognition target voice of the second voice recognition engine 43 is set to the voice input from the audio source 6 (step 502).

Then, the hit data notification is generated from the first voice recognition engine 41 (step 504), the hit data notification is generated from the second voice recognition engine 43 (step 506), and the timer timeout occurs (step 508). ) And monitor.

Then, if the notification of the hit data from the second voice recognition engine 43 occurs (step 506), the mask flag is set (step 514), and the number indicated by the hit data notified from the second voice recognition engine 43 (step 514). No.) is set as the adjustment word number (step 516). Then, the timer described above is started by setting a predetermined timeout time (step 518), and the process returns to the monitoring of steps 504, 506, and 508.

On the other hand, when the timer time-out occurs (step 508), the mask flag is cleared (step 510), and the adjustment word number setting is cleared (step 512). Then, the process returns to the monitoring of steps 504, 506, and 508.

Further, when the notification of the hit data from the first voice recognition engine 41 occurs (step 504), it is examined whether or not the mask flag is set (step 520), and if the mask flag is not set, the first step. 1 The word of the number (No.) indicated by the hit data from the first speech recognition engine 41 in the speech recognition dictionary is calculated as the word to be the recognition result, and the calculated recognition result is output to the data processing unit 1 (step). 524). Then, the audio cancel mode recognition process is terminated.

On the other hand, when the mask flag is set in step 520, it is checked whether or not the number (No.) indicated by the hit data from the first voice recognition engine 41 and the adjustment word number match (step 522). If so, the process returns to the monitoring of steps 504, 506, and 508.

On the other hand, when the word of the number (No.) indicated by the hit data from the first speech recognition engine 41 does not match the adjustment word (step 522), the first speech recognition engine 41 of the first speech recognition dictionary The word of the number (No.) indicated by the hit data from is calculated as the word to be the recognition result, and the calculated recognition result is output to the data processing unit 1 (step 524). Then, the audio cancel mode recognition process is terminated.

The audio cancel mode recognition process performed by the recognition adjustment unit 45 when the audio cancel mode is set has been described above.
Here, according to such an audio cancel mode recognition process, the word for which the second voice recognition engine 43 detects a hit with respect to the voice input from the audio source 6 is subsequently the first voice recognition engine 41 for a certain period of time. Even if a hit is detected in, it is not recognized as a recognition result.

Further, when the voice output by the audio source 6 wraps around the microphone 3 and a word hit is detected by the first voice recognition engine 41 for the voice, the threshold Th of the word is the first voice recognition. Since the second voice recognition dictionary 44 is set larger than the dictionary 42, and the voice of the audio source 6 input to the second voice recognition engine 43 has better voice quality, the second voice is set before that. The recognition engine 43 detects the hit of the word.

Therefore, even if the voice output by the audio source 6 wraps around the microphone 3 and a word hit is detected by the first voice recognition engine 41 for the voice, the word is recognized as a recognition result by the data processing unit. It is suppressed that it is output to 1.

Next, the similar word strong identification mode recognition process performed by the recognition adjustment unit 45 when the similar word strong identification mode is set will be described.
FIG. 6 shows the procedure of the similar word strong identification mode recognition process.
As shown in the figure, in the similar word strong identification mode recognition process, the recognition adjustment unit 45 first sets the recognition target voice of the second voice recognition engine 43 to the voice input from the microphone 3 (step 602).

Then, the generation of the hit data notification from the first voice recognition engine 41 or the second voice recognition engine 43 is monitored (step 604).
Then, when the notification of the hit data from the first voice recognition engine 41 or the second voice recognition engine 43 occurs (step 604), the notified hit data is saved in association with the notification source voice recognition engine (step 604). Step 606), start a timer with a predetermined timeout period (eg, 500 ms) set (step 608).

Next, the occurrence of hit data notification from the first voice recognition engine 41 or the second voice recognition engine 43 (step 610) and the occurrence of the timer timeout started in step 608 (step 612) are monitored. To do.

Then, when the notification of the hit data from the first voice recognition engine 41 or the second voice recognition engine 43 occurs (step 610), the notified hit data is saved in association with the notification source voice recognition engine (step 610). Step 618), return to monitoring in steps 610 and 612.

On the other hand, when the timer time-out occurs (step 612), the word whose stored hit data indicates the number (No.) is analyzed, and the word to be the recognition result is calculated (step 614).

Here, in step 614, the word to be the recognition result is calculated as follows.
That is, first, for each hit data stored in association with the voice recognition engine of the notification source, the word number (No.) represented by the hit data and the voice recognition engine of the notification source are used in the first voice recognition engine 41. A word registered in the first speech recognition dictionary 42 in which a hit is detected is distinguished from a word registered in the second speech recognition dictionary 44 in which a hit is detected by the second speech recognition engine 43.

Then, the value of the word number (No.) registered in the first voice recognition dictionary 42 in which the hit was detected by the first voice recognition engine 41 and the second value in which the hit was detected by the second voice recognition engine 43. The value that appears most frequently is calculated as one of the value of the reference number (ref) of the word registered in the voice recognition dictionary 44.

Then, if the value that appears most frequently is one, the word registered in the first speech recognition dictionary 42 having that value as the number (No.) is calculated as the word as the recognition result.
That is, for example, the third layer recognition data in FIG. 2 is the current recognition data, the main speech recognition dictionary of the third layer recognition data is set in the first speech recognition dictionary 42, and the sub speech of the third layer recognition data. When the recognition dictionary is set in the second speech recognition dictionary 44, the word registered in the first speech recognition dictionary 42 in which a hit is detected by the first speech recognition engine 41 has a number (No.) of 1. "I'm going to the first" and the number (No.) is 5 "I'm going to the first", and the word registered in the second speech recognition dictionary 44 where a hit was detected by the second speech recognition engine 43 is referenced. If the number (ref) is 1 for "person" and the reference number (ref) is 1, "go to the first", the number of occurrences of value 1 is 3, the number of appearances of value 5 is 1, and the value is 1. Is the value that appears most often, so the word "Ichime ni Iku" registered in the first speech recognition dictionary 42, which has 1 as the number (No.), is calculated as the recognition result.

By doing so, multiple words with similar pronunciations are comprehensively evaluated in consideration of evaluation by different thresholds and evaluation of consistency with spoken voice focusing only on the part whose pronunciation is different from other words. As a result, it becomes possible to perform good speech recognition even for a plurality of words having similar pronunciations.

On the other hand, if there are a plurality of values that appear most frequently, the value that appears most frequently among the words registered in the second speech recognition dictionary 44 in which a hit is detected by the second speech recognition engine 43 is referred to as a reference number (ref). ), And the value (Y) is registered in the partial flag (P), that is, the word is set to be the part of the word registered in the first speech recognition dictionary 42. Checks if the most frequently occurring partial word that is the word that is being used exists.

Then, if the most appearing partial word exists and the reference numbers (ref) of the most appearing partial words that existed are all equal, the first speech recognition has the value of the reference number (ref) of the existing word as the number (No.). The word registered in the dictionary 42 is calculated as the word as the recognition result.

That is, for example, the third layer recognition data in FIG. 2 is the current recognition data, the main speech recognition dictionary of the third layer recognition data is set in the first speech recognition dictionary 42, and the sub speech of the third layer recognition data. When the recognition dictionary is set in the second speech recognition dictionary 44, the word registered in the first speech recognition dictionary 42 in which a hit is detected by the first speech recognition engine 41 has a number (No.) of 1. "Go to the first" and "Go to the second" with the number (No.) 2 and "Go to the second" with the number (No.) 5 and the hit is detected by the second voice recognition engine 43. If the words registered in the second speech recognition dictionary 44 are "person" with a reference number (ref) of 1 and "go to tsutsume" with a reference number (ref) of 5, 1 and 5 are the most. Although it is a value that appears frequently, a value 1 or 5 that is a word registered in the second speech recognition dictionary 44 in which a hit is detected by the second speech recognition engine 43 and appears most frequently as a reference number (ref) is used. Among the words that have, the word of the second speech recognition dictionary 44 in which the value (Y) is registered in the partial flag (P) is only the "person" whose reference number (ref) is 1, so 1 is the number (1). The word "first go", which is registered in the first speech recognition dictionary 42 and has as No.), is calculated as the recognition result.

By doing so, it becomes possible to calculate the word as the recognition result by giving more importance to the evaluation result of the consistency with the uttered voice focusing only on the part whose pronunciation is different from other words.

On the other hand, when the most frequently appearing partial word does not exist, the word registered in the first speech recognition dictionary 42 having the most frequently appearing value as the number (No.) and the most frequently appearing value are referred to as reference numbers (ref). ), Among the words registered in the second speech recognition dictionary 44 in which a hit is detected by the second speech recognition engine 43, the word having the smallest threshold Th of the word is calculated and calculated. If is a word of the first speech recognition dictionary 42, the word is calculated as a word that is the recognition result, and if the calculated word is a word of the second speech recognition dictionary 44, the reference number (ref) of the word is calculated. A word registered in the first speech recognition dictionary 42 having a value as a number (No.) is calculated as a word as a recognition result.

Further, if the most appearing partial word exists and the reference numbers (ref) of the most appearing partial words that existed are not all equal, the reference number (ref) of the most appearing partial word is used as the reference number (ref). 2 The first voice recognition engine 41 having a word registered in the second voice recognition dictionary 44 in which a hit is detected by the voice recognition engine 43 and a reference number (ref) of the most frequently appearing partial word as a number (No.). Among the words registered in the 1 speech recognition dictionary in which a hit is detected, the word having the minimum threshold Th of the word is calculated, and if the calculated word is the word of the 1st speech recognition dictionary 42, The word is calculated as a word as a recognition result, and if the calculated word is a word in the second speech recognition dictionary 44, the first speech has the value of the reference number (ref) of the word as the number (No.). The word registered in the recognition dictionary 42 is calculated as the word as the recognition result.

However, in this step 614, when there is only one word registered in the first speech recognition dictionary 42 in which a hit is detected by the first speech recognition engine 41, that word is calculated as a word as a recognition result. Then, when there are a plurality of words registered in the first speech recognition dictionary 42 in which a hit is detected by the first speech recognition engine 41, the word number (No.) is assigned to each of the plurality of words. It is checked whether or not the hit of the word of the second speech recognition dictionary 44 in which the value (Y) is registered in the partial flag (P) as the reference number (ref) is detected by the second speech recognition engine 43, and the hit is found. When only one word detected by the second voice recognition engine 43 exists among the plurality of words, the process may be performed to calculate the word as the recognition result.

Then, the calculated recognition result is output to the data processing unit 1 (step 616), and the similar word strong identification mode recognition process is terminated.
The similar word strong identification mode recognition process performed by the recognition adjustment unit 45 when the similar word strong identification mode is set has been described above.
Here, an operation example of the above-mentioned information processing apparatus is shown in FIG.
First, the activated data processing unit 1 starts information processing for displaying a guide map for car navigation on the display device 7 and information processing for outputting music from the audio source 6.
Further, when the data processing unit 1 is activated, the voice input setting process is started, and the first layer recognition data is set as the current recognition data, so that the main voice of the first layer recognition data is set as shown in FIG. 7a2. The recognition data is set in the first speech recognition dictionary 42, the sub-speech recognition data of the first layer recognition data is set in the second voice recognition dictionary 44 as shown in FIG. 7a3, and the recognition adjustment unit 45 is in the first layer. Set the audio cancel mode registered as the recognition mode in the recognition data.

As a result, in the recognition adjustment unit 45 of the voice input unit 4, the voice recognition of the word registered in the first voice recognition dictionary 42 by the audio cancel mode recognition process causes erroneous recognition by the output voice of the audio source 6. 2 It is performed while suppressing using the voice recognition dictionary 44.

Next, in this state, when the user utters the word "Mokukichisete" registered in the first speech recognition dictionary 42, the recognition adjustment unit 45 uses the word "Mokukichisete" as the recognition result. It is calculated, and the word "Mokukichisetei" is output from the voice input unit 4 to the data processing unit 1 as a recognition result.

If the data processing unit 1 accepts the word "Mokukichisetei" as a recognition result, it responds to this and corresponds to the word "Mokukichisete" in the main speech recognition dictionary of the first layer recognition data. The second layer recognition data to be performed is set as the current recognition data. Then, as shown in FIG. 7b1, the display screen of the display device 7 is changed to a screen including a list of words registered in the main voice recognition data of the second layer recognition data set in the current recognition data.

Further, the data processing unit 1 sets the main voice recognition data of the second layer recognition data set in the current recognition data as shown in FIG. 7b2 in the first voice recognition dictionary 42, and sets the current recognition data as shown in FIG. 7b3. The sub-speech recognition data of the second layer recognition data set in is set in the second voice recognition dictionary 44, and is registered in the recognition adjustment unit 45 in the second layer recognition data set in the current recognition data as a recognition mode. Set the audio cancel mode.

As a result, in the recognition adjustment unit 45 of the voice input unit 4, the voice recognition of the word registered in the first voice recognition dictionary 42 by the audio cancel mode recognition process causes erroneous recognition by the output voice of the audio source 6. 2 It is performed while suppressing using the voice recognition dictionary 44.

Next, in this state, when the user utters the word "Chikaku no Ramenya" registered in the first speech recognition dictionary 42, the recognition adjustment unit 45 calculates the word "Chikaku no Ramenya" as the recognition result. , The word "Chikaku no Ramenya" is output from the voice input unit 4 to the data processing unit 1 as a recognition result.

When the data processing unit 1 receives the word "Chikaku no Ramenya" as the recognition result, in response to this, the data processing unit 1 responds to the word "Chikaku" in the main speech recognition dictionary of the second layer recognition data set in the current recognition data. The third layer recognition data corresponding to "Noramenya" is set as the current recognition data. Then, the ramen shop near the current position is searched by the car navigation function provided in the data processing unit 1, and as shown in FIG. 7c1, the display screen of the display device 7 is set to the current recognition data in the third layer. The list of words registered in the main speech recognition data of the recognition data is set in the current recognition data in the list. From "I go to the first" registered in the main speech recognition data of the third layer recognition data, "when Change the screen to include the display of the five ramen shops near the searched current position in the corresponding form in the five words of "Tsume ni Iku".

Further, the data processing unit 1 sets the main voice recognition data of the third layer recognition data set in the current recognition data as shown in FIG. 7c2 in the first voice recognition dictionary 42, and sets the current recognition data as shown in FIG. 7c3. The sub-speech recognition data of the third layer recognition data set in is set in the second voice recognition dictionary 44, and is registered as a recognition mode in the third layer recognition data set in the current recognition data in the recognition adjustment unit 45. Set the word strong identification mode.

As a result, in the recognition adjustment unit 45 of the voice input unit 4, different thresholds using the first voice recognition dictionary 42 and the second voice recognition dictionary 44 are performed by the similar word strong identification mode recognition process as described above. A form that better identifies multiple words with similar pronunciation by comprehensive evaluation that takes into consideration the evaluation by and the evaluation of consistency with the spoken voice focusing only on the part whose pronunciation is different from other words. Then, the voice recognition of the word registered in the first voice recognition dictionary 42 is performed.

Then, in this state, when the user utters the word "first go" registered in the first speech recognition dictionary 42, the recognition adjustment unit 45 calculates the word "first go" as a recognition result. The word "first go" is output from the voice input unit 4 to the data processing unit 1 as a recognition result.

If the data processing unit 1 accepts the word "go to the first" as a recognition result, it sets the ramen shop shown on the screen of FIG. 7c1 as the destination in association with the word "go to the first" and data. In the car navigation function provided in the processing unit 1, processing for directions to the destination is started.

By the way, when the similar word strong identification mode is set as described above, the suppression using the second speech recognition dictionary 44 of the erroneous recognition by the output speech of the audio source 6 is not performed. However, the similar word strong identification mode is set when the recognition data of the layer other than the first layer recognition data is set as the current recognition data, and the recognition data of the layer other than the first layer recognition data is set. When set to the active recognition data, it is when the user is continuously performing a series of hierarchical voice inputs.

Therefore, the period from the setting of the similar word strong identification mode to the voice input by the user's utterance is short, and during this period, the voice of the same word as the word set in the first voice recognition dictionary 42 is heard. It is rarely output from the audio source 6.

Therefore, when the recognition data of the layer other than the first layer recognition data is set as the current recognition data, it is not necessary to suppress the false recognition by the output voice of the audio source 6 by using the second voice recognition dictionary 44. Practically, there is no problem.

When the first layer recognition data is set as the current recognition data, the period from when the first layer recognition data is set to the current recognition data until the voice input by the user's utterance occurs is indefinite, and during this period In addition, there is a high possibility that the voice of the same word as the word set in the first voice recognition dictionary 42 is output from the audio source 6. Therefore, when the first layer recognition data is set as the current recognition data, it is important to suppress the false recognition by the output voice of the audio source 6 by using the second voice recognition dictionary 44.

The embodiment of the present invention has been described above.
In the above embodiment, the sub-speech recognition dictionary of the recognition data in which the recognition mode is set to the similar word strong identification mode is registered in the main speech recognition dictionary, which is registered in the main speech recognition dictionary of the recognition data. Words that match each word that has similar pronunciation to other words that are used, or parts that do not have similar pronunciation to other words in the word are registered as words, but the recognition mode is set to similar word strong identification mode. In the sub-speech recognition dictionary of the recognized data, only words and phrases that match each word whose pronunciation is similar to other words registered in the main speech recognition dictionary of the recognition data are registered. Or, only the part whose pronunciation is not similar to that of other words of each word whose pronunciation is similar to that of other words registered in the main speech recognition dictionary may be registered.

Further, in the above embodiment, the recognition modes of the recognition data of each layer other than the first layer recognition data may be set to the similar word strong identification mode.

1 ... Data processing unit, 2 ... Dictionary DB, 3 ... Microphone, 4 ... Voice input unit, 5 ... Speaker, 6 ... Audio source, 7 ... Display device, 8 ... Peripheral device, 41 ... First voice recognition engine, 42 ... 1st voice recognition dictionary, 43 ... 2nd voice recognition engine, 44 ... 2nd voice recognition dictionary, 45 ... recognition adjustment unit.

Claims (10)

It is a voice recognition system that recognizes the voice spoken by the user.
With a microphone
A first speech recognition dictionary in which a plurality of first recognition candidates that are words are registered, and
The voice picked up by the microphone is input, and among a plurality of first recognition candidates registered in the first voice recognition dictionary, the first recognition candidate that matches the input voice better than a predetermined level is detected as a target candidate. The first voice recognition means to be
Second voice recognition means and
A second speech recognition dictionary used by the second speech recognition means,
Has a means of recognition
The speech recognition system has one or more recognition modes, including a first recognition mode.
In the second speech recognition dictionary used by the second speech recognition means in the first recognition mode, a plurality of words having partially common pronunciations registered as first recognition candidates in the first speech recognition dictionary. Each word excerpt part, which is a part whose pronunciation is not common to other words, is registered as a second recognition candidate.
In the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone, and selects the input voice from a plurality of second recognition candidates registered in the second voice recognition dictionary. A second recognition candidate that matches better than a predetermined level is detected as a target candidate, and
In the first recognition mode, the recognition means is the target of each word registered as a first recognition candidate in the first speech recognition dictionary by the first speech recognition means and the second speech recognition means. Calculate the number of the first recognition candidate, which is the word, and the second recognition candidate, which is an excerpt of the word, detected as candidates, and recognize the word with the largest calculated number as the word spoken by the user. A voice recognition system featuring. It is a voice recognition system that recognizes the voice spoken by the user.
With a microphone
A plurality of first recognition candidates that are words, a first speech recognition dictionary registered with a plurality of first threshold values corresponding to each first recognition candidate, and
The voice picked up by the microphone is input, and the voice that is better than the level indicated by the corresponding first threshold value is matched from among the plurality of first recognition candidates registered in the first voice recognition dictionary. 1 First voice recognition means for detecting recognition candidates as target candidates,
Second voice recognition means and
A second speech recognition dictionary used by the second speech recognition means,
Has a means of recognition
The speech recognition system has one or more recognition modes, including a first recognition mode.
In the second voice recognition dictionary used by the second voice recognition means in the first recognition mode, the same words as the plurality of words registered as the first recognition candidates in the first voice recognition dictionary are each second. As recognition candidates, they are registered together with a plurality of second threshold values corresponding to each of the second recognition candidates.
In the first recognition mode, the second threshold value corresponding to each second recognition candidate registered in the second speech recognition dictionary is the same word as the word registered as the second recognition candidate. It shows a better level of matching than the first threshold corresponding to the recognition candidate.
In the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone, and the corresponding second recognition candidate is selected from the plurality of second recognition candidates registered in the second voice recognition dictionary. The second recognition candidate that matches the input voice better than the level indicated by the threshold is detected as the target candidate.
In the first recognition mode, the recognition means is the target of each word registered as a first recognition candidate in the first speech recognition dictionary by the first speech recognition means and the second speech recognition means. A voice recognition system characterized in that the number of the first recognition candidate and the second recognition candidate, which are the words detected as candidates, is calculated, and the word with the largest calculated number is recognized as the word spoken by the user. .. It is a voice recognition system that recognizes the voice spoken by the user.
With a microphone
A plurality of first recognition candidates that are words, a first speech recognition dictionary registered with a plurality of first threshold values corresponding to each first recognition candidate, and
The voice picked up by the microphone is input, and the voice that is better than the level indicated by the corresponding first threshold value is matched from among the plurality of first recognition candidates registered in the first voice recognition dictionary. 1 First voice recognition means for detecting recognition candidates as target candidates,
Second voice recognition means and
The second voice recognition dictionary used by the second voice recognition means and
Has a means of recognition
The speech recognition system has one or more recognition modes, including a first recognition mode.
In the second speech recognition dictionary used by the second speech recognition means in the first recognition mode, a plurality of words having a partially common pronunciation registered as the first recognition candidate in the first speech recognition dictionary The word excerpt part, which is a part whose pronunciation is not common to other words, and the same word as a plurality of words registered as the first recognition candidate in the first speech recognition dictionary are the second recognition candidates, respectively. It is registered with a plurality of second threshold values corresponding to each second recognition candidate.
In the first recognition mode, the second threshold value corresponding to the second recognition candidate, which is the same word as the word registered as the first recognition candidate, registered in the second speech recognition dictionary is the said. It shows a better level of matching than the first threshold value corresponding to the first recognition candidate, which is the same word as the word registered as the second recognition candidate.
In the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone, and the corresponding second recognition candidate is selected from the plurality of second recognition candidates registered in the second voice recognition dictionary. The second recognition candidate that matches the input voice better than the level indicated by the threshold is detected as the target candidate.
In the first recognition mode, the recognition means is the target of each word registered as a first recognition candidate in the first speech recognition dictionary by the first speech recognition means and the second speech recognition means. The number of the first recognition candidate that is the word, the second recognition candidate that is the word, and the second recognition candidate that is the excerpt part of the word that are detected as candidates is calculated, and the word with the largest calculated number is calculated. A voice recognition system characterized in that it is recognized as a word spoken by the user. The voice recognition system according to claim 3.
In the first recognition mode, when there are a plurality of words having the largest calculated number, the recognition means is a second excerpt portion of the words having the largest calculated number. A voice recognition system characterized in that a recognition candidate recognizes a word detected as the target candidate as a word spoken by a user. The voice recognition system according to claim 4.
In the first recognition mode, the recognition means has a plurality of words having the maximum calculated number, and among the words having the maximum calculated number, the second recognition candidate which is an excerpt portion of the word is When there are a plurality of words detected as the target candidates, the calculated number is the largest word, and the second recognition candidate, which is an excerpt part of the word, is among the words detected as the target candidates. The first threshold value corresponding to the first recognition candidate which is the word detected as the target candidate, and the second threshold value corresponding to the second recognition candidate which is the word detected as the target candidate. The feature is that the word whose minimum value with the second threshold value corresponding to the second recognition candidate, which is an excerpt portion of the word detected as the target candidate, is the smallest is recognized as the word spoken by the user. Voice recognition system. It is a voice recognition system that recognizes the voice spoken by the user.
With a microphone
A first speech recognition dictionary in which a plurality of first recognition candidates that are words are registered, and
The voice picked up by the microphone is input, and among a plurality of first recognition candidates registered in the first voice recognition dictionary, the first recognition candidate that matches the input voice better than a predetermined level is detected as a target candidate. The first voice recognition means to be
Second voice recognition means and
A second speech recognition dictionary used by the second speech recognition means,
Has a means of recognition
The speech recognition system has one or more recognition modes, including a first recognition mode.
In the second speech recognition dictionary used by the second speech recognition means in the first recognition mode, a plurality of words having a partially common pronunciation registered as first recognition candidates in the first speech recognition dictionary. Each word excerpt part, which is a part whose pronunciation is not common to other words, is registered as a second recognition candidate.
In the first recognition mode, the second voice recognition means inputs the voice picked up by the microphone, and selects the input voice from a plurality of second recognition candidates registered in the second voice recognition dictionary. A second recognition candidate that matches better than a predetermined level is detected as a target candidate, and
In the first recognition mode, when there are a plurality of words that are the first recognition candidates detected as the target candidates by the first voice recognition means, the recognition means among the plurality of words. A voice recognition system characterized in that a second recognition candidate, which is an excerpt portion of the word, recognizes a word detected as the target candidate by the second voice recognition means as a word spoken by a user. The voice recognition system according to claim 1, 2, 3, 4, 5 or 6.
The microphone is arranged in a space where the sound output from the audio source device is radiated from the speaker to the speaker.
The voice recognition mode has a plurality of recognition modes including the first recognition mode and the second recognition mode.
In the second voice recognition dictionary used by the second voice recognition means in the second recognition mode, the same words as the plurality of words registered as the first recognition candidates in the first voice recognition dictionary are each second. It is registered as a recognition candidate and
In the second recognition mode, the second voice recognition means inputs the voice output to the speaker by the audio source device, and is selected from a plurality of second recognition candidates registered in the second voice recognition dictionary. A second recognition candidate that matches the input voice better than a predetermined level is detected as a target candidate.
In the second recognition mode, the recognition means is a second recognition candidate detected as the target candidate during a predetermined period after the second recognition candidate is detected as the target candidate by the second voice recognition means. Even if the first recognition candidate which is the same word as is detected as the target candidate by the first voice recognition means, the word which is the first recognition candidate detected as the target candidate is recognized as the word spoken by the user. A voice recognition system characterized in that a word that is a first recognition candidate detected as a target candidate by the first voice recognition means is recognized as a word spoken by a user while suppressing this. The voice recognition system according to claim 7.
A storage means for storing a plurality of voice recognition data representing the first voice recognition dictionary and the second voice recognition dictionary, and
Voice input receiving means that accepts user's voice input and
It has a recognition mode control means for controlling the recognition mode, and has
The voice input receiving means sets the first voice recognition dictionary and the second voice recognition dictionary represented by the predetermined voice recognition data, and receives the word recognized by the recognition means as the voice input, and the first state. The word recognized by the recognition means while updating the voice recognition dictionary and the second voice recognition dictionary to the first voice recognition dictionary and the second voice recognition dictionary represented by the voice recognition data determined according to the word received as the previous voice input. Has a voice input sequence executing state that executes a sequence that accepts the voice input once or a plurality of times, and if a word is received as a voice input in the standby state, the state transitions to the voice input sequence executing state, and the above When the execution of the sequence in the voice input sequence execution state is completed, the state transitions to the standby state.
The recognition mode control means sets the second recognition mode as the recognition mode when the voice input receiving means is in the standby state, and when the voice input receiving means is in the voice input sequence executing state. A voice recognition system characterized in that the recognition mode is switched between the first recognition mode and the second recognition mode according to voice recognition data determined according to a word received as the previous voice input. .. The voice recognition system according to claim 7.
A storage means for storing a plurality of voice recognition data representing the first voice recognition dictionary and the second voice recognition dictionary, and
Voice input receiving means that accepts user's voice input and
It has a recognition mode control means for controlling the recognition mode, and has
The voice input receiving means sets the first voice recognition dictionary and the second voice recognition dictionary represented by the predetermined voice recognition data, and receives the word recognized by the recognition means as the voice input, and the first state. The word recognized by the recognition means while updating the voice recognition dictionary and the second voice recognition dictionary to the first voice recognition dictionary and the second voice recognition dictionary represented by the voice recognition data determined according to the word received as the previous voice input. Has a voice input sequence executing state that executes a sequence that accepts the voice input once or a plurality of times, and if a word is received as a voice input in the standby state, the state transitions to the voice input sequence executing state, and the above When the execution of the sequence in the voice input sequence execution state is completed, the state transitions to the standby state.
The recognition mode control means sets the second recognition mode as the recognition mode when the voice input receiving means is in the standby state, and when the voice input receiving means is in the voice input sequence executing state. , A voice recognition system characterized in that the first recognition mode is set as the recognition mode. The voice recognition system according to claim 1, 2, 3, 4, 5, 6, 7, 8 or 9.
The voice recognition system is a voice recognition system used for voice input in an in-vehicle system mounted on an automobile.