JP2009015148A - Speech recognition device, speech recognition method and speech recognition program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech recognition device with user friendliness, without spoiling easy input which is originally facilitated by speech input itself, when a correct word is input by speech again. <P>SOLUTION: The speech recognition device includes: a storage section for storing a word data; a speech recognition section in which a recognition score for expressing a matching degree of comparison is calculated by comparing the word data expressed by input speech with the word data stored in the storage section, and on the basis of the calculated recognition score, the recognized word is output; a receiving period determination means for determining a receiving period when re-input by speech can be received, on the basis of the recognition results calculated by the speech recognition section; and a speech re-recognition section for recognizing the word which is input again within the receiving period determined by the receiving period determination section, and the word which is recognized again is output. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a voice recognition apparatus, a voice recognition method, and a voice recognition program, and more specifically, a voice recognition apparatus, a voice recognition method, and a voice that correct a misrecognized utterance content to an utterance content to be recognized. It relates to recognition programs.

  In a conventional speech recognition apparatus, when a user utters a word, a word (hereinafter referred to as a word in the dictionary) stored in a dictionary held in advance and a word (hereinafter referred to as an utterance word) that are spoken are stored. Comparison is made using probability statistical processing or the like, and a probable word is notified to the user as a recognition result. At this time, there may occur a case where the word spoken by the user is different from the informed word (hereinafter, such phenomenon is referred to as misrecognition). There are various causes for this, but it is very difficult to recognize 100% correctly.

  On the other hand, in speech recognition processing using probability statistical processing, when comparing an utterance word with a word in the dictionary, a confidence level (word reliability, which indicates how much the utterance word is acoustically similar to the word in the dictionary) May be called likelihood etc.).

Using such confidence level, if the confidence level is 90%, the word is accepted as correctly recognized, and if the confidence level is about 50%, the user is inquired about the recognized word, and the correct word is again displayed. (For example, refer to Patent Document 1).
JP 2001-175276 A

  However, in the conventional voice recognition, when the degree of confidence is low, the user is inquired about the recognized word, and the user does not need to perform an operation such as pressing a re-input button again to input the correct word again. Must not.

  For this reason, the correct word cannot be input again without further one-step operation such as pressing the re-input button or the like, and the ease of input inherent in the voice input itself is impaired. It lacks user convenience.

  Therefore, the present invention has been made in view of the above problems. That is, an object of the present invention is to provide a speech recognition device, a speech recognition method, and a speech recognition program that are rich in user convenience without impairing the ease of input inherent in speech input itself when a correct word is input again. And

  The first aspect of the present invention is directed to a voice recognition device. The present invention compares a storage unit that stores word data, word data represented by input speech and word data stored in the storage unit, and calculates a recognition score that represents a matching degree at the time of comparison. A speech recognition unit that outputs the recognized word based on the recognition score, a reception time determination unit that determines a reception time that can accept re-input by speech based on the recognition score calculated by the speech recognition unit, and a reception A speech re-recognition unit that re-recognizes a word represented by the speech re-input within the reception time determined by the time determination unit and outputs the re-recognized word.

  With this configuration, it is possible to provide a speech recognition device that is rich in user convenience without impairing the ease of input inherent in speech input itself when inputting correct words again.

  In addition, it is preferable to further include an acceptance time notification unit that notifies the user by a display object that changes with a change in the acceptance time determined by the acceptance time determination unit.

  According to this configuration, since the time when the user can correctly change the recognition result is notified according to the degree of confidence of recognition on the device side, the user can correct the erroneous recognition without performing the correction operation.

  Moreover, it is preferable that the reception time notification unit stops the change of the display object when the voice is re-input within the reception time determined by the reception time determination unit.

  With this configuration, the user can intuitively understand that the device side is processing (recognizing) the content spoken by itself.

  Moreover, it is preferable that an acceptance time alerting | reporting part changes the position on the display area of the display object showing the acceptance time which the acceptance time determination part determined with the change of the said acceptance time.

  With this configuration, it is possible to intuitively grasp the remaining time that can be input again only by looking at the distance between the position where the movement of the object starts and the position where it ends.

  Moreover, it is preferable that the reception time notification unit changes the amount of the scale of the display object representing the reception time determined by the reception time determination unit together with the change in the reception time.

  With this configuration, the user can intuitively grasp the urgency of the time to be corrected.

  Moreover, it is preferable that an acceptance time alerting | reporting part changes the magnitude | size of the display object showing the acceptance time which the acceptance time determination part determined with the change of the said acceptance time.

  With this configuration, the user can intuitively estimate the degree of confidence on the device side from the smallness of the first character.

  Moreover, it is preferable that an acceptance time alerting | reporting part changes the transparency of the display object showing the acceptance time which the acceptance time determination part determined with the change of the said acceptance time.

  With this configuration, the user can intuitively estimate the degree of confidence on the device side from the transparency of the first character.

  In addition, it is preferable that the voice re-recognition unit removes words recognized by the voice recognition unit from the storage unit and re-recognizes during the time determined by the reception time notification unit.

  With this configuration, even when a voice is input to correct the result recognized this time, the same word as the previous one is not recognized again, and the recognition rate is increased.

  The speech re-recognition unit preferably re-recognizes a word in which a predetermined word is added to the word recognized by the speech recognition unit during the time determined by the reception time determination unit.

  With this configuration, when a user tries to utter a word to correct the result recognized this time, he / she is upset when he / she sees an incorrect result and utters an unnecessary word along with the word. Can output a desired word as a recognition result.

  The second aspect of the present invention is directed to a speech recognition method. The present invention compares the word data represented by the input speech with the word data stored in the storage unit, calculates a recognition score representing the matching degree at the time of comparison, and recognizes based on the calculated recognition score A speech recognition step for outputting a word, a reception time determination step for determining a reception time for accepting re-input by voice based on the recognition score calculated in the speech recognition step, and a reception time determined in the reception time determination step A speech re-recognition step of re-recognizing the word represented by the speech re-input to and outputting the re-recognized word.

  With this configuration, when a correct word is input again by voice, it is possible to provide a voice recognition method that is rich in user convenience without impairing the ease of input inherent in the voice input itself.

  The third aspect of the present invention is directed to a voice recognition program executed by a computer of a voice recognition device. The present invention compares the word data represented by the input speech with the word data stored in the storage unit in a computer, calculates a recognition score representing the matching degree at the time of comparison, and based on the calculated recognition score The speech recognition step that outputs the recognized word, the reception time determination step that determines the reception time that can accept re-input by voice based on the recognition score calculated in the speech recognition step, and the reception time determination step A speech re-recognition step of re-recognizing a word represented by the speech re-input within the reception time and outputting the re-recognized word is executed.

  With this configuration, when a correct word is input again by voice, it is possible to provide a voice recognition program which is rich in user convenience without impairing the ease of input inherent in the voice input itself.

  As described above, according to each aspect of the present invention, when a correct word is input again by speech, the speech recognition device and the speech that are rich in user convenience without impairing the input ease inherent in the speech input itself. A recognition method and a speech recognition program can be provided.

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

  FIG. 1 is a block diagram showing the overall configuration of a speech recognition apparatus according to an embodiment of the present invention. In FIG. 1, the speech recognition apparatus includes a speech input unit 100, a recognition unit 200, a recognition target word storage unit 300, a correction timing control unit 400, a notification content control unit 500, and a notification unit 600. The voice input unit 100 further includes a recognition start / end unit 110, and the recognition unit 200 further includes a recognition score calculation unit 210.

  The voice input unit 100 is, for example, a microphone that captures voice spoken by the user. When capturing the voice, the user operates the recognition start / end unit 110 that notifies the recognition unit 200 that voice data will be input in advance. When operated by the user, the recognition start / end unit 110 operates, for example, by a button operation or by uttering a special word.

  Upon receiving a notification from the recognition start / end unit 110, the recognition unit 200 starts taking in the input voice data, and the words stored in the recognition target word storage unit 300 (hereinafter referred to as words in the dictionary). Compared with the above, words that are acoustically close among the words in the dictionary are extracted. As a technique for realizing the above comparison process, a probability statistical processing technique such as HMM (Hidden Markov Model) applied in many speech recognition processes may be applied. When this acoustically close word is extracted, the recognition score calculation unit 210 calculates how close the words are (similar) to each other as a score. As such a score, general likelihood, word reliability, etc. may be used. In the present embodiment, this score is assumed to be 0 in the range from 0 to 10 and 0 if it is closest, and 10 if it is far.

  Here, each process will be described by taking as an example the time when the user utters the word “see”.

  The recognizing unit 200 receives the voice data of “see”, and extracts words spoken of “see” from the words in the dictionary stored in the recognition target word storage unit 300. At this time, it is assumed that the words “see” and “map” are registered in the recognition target word storage unit 300 in advance, and the recognition unit 200 extracts “map” as the acoustically closest word and recognizes the result. And Assume that the score calculated by the recognition score calculation unit 210 for the recognized word “map” is “6”.

  Here, the flow of processing will be described with reference to FIG. The recognition unit 200 transmits the recognition word (“map”) and the calculated score (“6”) to the correction timing control unit 400 (step S200).

  The correction timing control unit 400 determines a time during which the recognized word can be corrected using the notified calculated score (step S201). For example, in this case, since the score is “6”, a timing of 6 seconds is set. However, this setting method shortens the time that can be corrected if there is confidence in the recognition result, and the time that can be corrected if there is no confidence. What is necessary is just to set so that it may become long. The time at this time may be a discrete value or a continuous value with respect to the score calculated by the recognition unit 200.

  Next, the correction timing control unit 400 notifies the recognition start / end unit 110 of a trigger to start recognition (step S203). Upon receiving this notification, the recognition start / end unit 110 notifies the recognition unit 200 of a start trigger so that the recognition process can be performed (step S204).

  Here, the process branches depending on whether the user speaks within the correctable time. That is, in the case of this example, it is whether the user has spoken a word within 6 seconds.

  When the user speaks within the time, the recognition unit 200 performs speech recognition processing with reference to the recognition target word storage unit 300, and extracts a recognition word (step S205, step S206). After that, the correction timing control unit 400 notified of the recognized word and score corrects the previously recognized word to the word notified this time and re-recognizes it.

  Thus, for example, if the user utters “see” again this time, and the recognition unit 200 recognizes “see”, the previously recognized word (“map”) is changed to the word (“see”) uttered this time. ) To correct the previous word and recognize it again. Therefore, when the user corrects and re-recognizes the last spoken word, the user does not need to operate the recognition start / end unit 110 or correct the previous recognition result, and the word to be recognized again Correction operations can be performed simply by speaking. And when the score is low (when the degree of confidence in recognition is low), the time that can be corrected is taken long, so that the user can be given many opportunities for correction. On the other hand, when the score is high (when the degree of confidence in recognition is low), the correction time is short, so the waiting time is short for the user and the next operation is not hindered.

  In step S202, the previous recognized word (in this example, the word “map”) is temporarily deleted from the recognition target word storage unit 300 referred to by the recognition unit 200. This is effective in that when the user speaks again, words that are acoustically similar are not recognized, so that erroneous recognition is less likely to occur and the recognition rate is increased.

  In step S 213, a word in which an unnecessary word such as “Ett” or “That” is added before each word stored in the recognition target word storage unit 300 is temporarily stored in the recognition target word. Part 300 is added. This is to prevent the recognition unit 200 from erroneously recognizing unnecessary words that are uttered when the user is notified of an unexpected recognition result.

  Then, whether or not to perform the processes in steps S202 and S213 may be determined based on the score. In particular, when the score calculated by the recognizing unit 200 is a highly reliable value, the processing load such as word registration is high and can be omitted to reduce the load.

  Although not shown in FIG. 2, when the recognition unit 200 finishes the recognition or the correctable time has passed after the processing of step S202 and step S213 is completed, the recognition target word storage unit 300 is displayed. In contrast, the contents processed in step S202 and step S213 are restored.

  Next, a case where there is no utterance from the user even after the correctable time has elapsed will be described. In the case of this example, this is a case where the voice data is not input from the voice input unit 100 even after 6 seconds. In this case, the correction timing control unit 600 starts recognition immediately after the correctable time elapses. The end unit 110 is notified of a trigger to end recognition. Then, the recognition start / end unit 110 receives this notification and notifies the recognition unit 200 of an end trigger, so that the recognition unit 200 does not perform a recognition process unless a start trigger by the next user operation is input.

  By processing in this way, even if the recognition score of the word spoken by the user is low, if the word is correctly recognized, the user does not need to perform any operation, so there is no need to perform an extra operation.

  On the other hand, the correction timing control unit 400 notifies the notification content control unit 500 of the calculated correctable time and the recognized word.

  The notification content control unit 500 controls the notification method to notify the user that the recognized word can be corrected, and notifies the notification unit 600 of the notification method. For example, the notification unit 600 reports a GUI display such as a liquid crystal display or an organic EL, a display capable of smoothing sound (sound, beep sound) such as a speaker, and a tactile sense such as a motor. The form etc. which can be done are considered. Here, as a notification method, the methods as shown in FIGS. 3A to 6 can be considered. In these examples, it is assumed that the recognition unit 200 recognizes a “map” and the correctable time is 6 seconds.

  3A and 3B show an example in which the movement amount changes.

  FIG. 3A and FIG. 3B show the change in the correctable time from state 1 (start) to state 3 (end) on the display D300 for notifying which outputs the authentication result. In this example, the state is divided into three states, but the state may change continuously.

  First, FIG. 3A shows a state in which the word W300 recognized by the recognition unit 200 is moving at a constant speed (or acceleration) in the display D300 during the correctable time. In this example, the correction time ends when the word W300 moves to the position indicated by the region D301, and the recognition result W300 is determined as the recognition result. In this example, the correction time is 6 seconds (the time required from state 1 to state 3). However, if the correction possible time is 3 seconds, the initial position of the word W300 is placed at the state 2 position. Can do. In other words, the user can grasp how much correction time remains after seeing the distance between the word W300 and the region D301.

  FIG. 3B is a diagram showing that the scale D303 increases as time passes and gradually fills the region D302. By displaying such a scale, the user can intuitively grasp the remaining correction time. Further, the width of the scale D303 or the region D302 may be changed according to the correction time. As a result, the correction time can be changed without changing the time to increase one scale. In addition, although the scale D303 is increased in the state 1 (start) in FIG. 3B, the scale D303 may be set to 0.

  Next, an example in which the size changes will be described with reference to FIG.

  In FIG. 4 as well, the state during the correctable time will be described using state 1 (start) to state 3 (end). In the state where the recognition unit 200 recognizes the word W300 in which correction can be started (state 1), the character size is reduced and gradually increased, and finally the correction time is set when it is the largest. finish. By doing so, the user can intuitively estimate the degree of confidence on the device side from the smallness of the first character, and can also understand that it takes a long time to correct. Moreover, since the change is on the spot, the amount of movement of the user's line of sight is small, and it is considered suitable for devices such as cars.

  Further, in the form as shown in FIG. 4, it may be possible to change not the size of characters but the speed of flashing characters of the word W300 (not shown).

  In such a case, the user sees the blinking speed by making the blinking speed (cycle) fast (short) in state 1 and then slowly slowing (longening) gradually and not blinking in state 3. Thus, the remaining time can be estimated, and the content (recognition result) most noticeable by the user blinks, and it can be suggested from the device side that there is a possibility of erroneous recognition. Note that the object whose speed is changed may be the size of the word, the moving speed of the word, or the like.

  Further, as shown in FIG. 5, the transparency for displaying the recognized word may be changed.

  Also in this case, the degree of confidence of the recognized word W300 can be intuitively presented to the user, and the remaining time that can be corrected can be estimated in a sense.

  In addition, as shown in FIG. 6, it is conceivable that the remaining time of the correctable time is displayed in the time display area D304.

  In this case, since the correctable time is directly displayed to the user, the remaining time is expressed very simply and is easy to understand.

  Note that the examples shown in FIGS. 3A to 6 described above may be displayed in organic combination.

  While making these changes, the user can utter the word he / she wants to correct when newly correcting the word. For example, when “speak” is spoken, the recognition unit 200 starts recognition without operating the recognition start / end unit 110. When the recognizing unit 200 starts the recognition process, the notification content control unit 500 stops various changes as shown in FIGS. 3A to 6.

  As a result, the user can intuitively understand that the device side processes (recognizes) the content of its own utterance when the change is eliminated. As other display methods, icons may be displayed or changed, the color of the screen may be changed, or guidance indicating the state with characters may be provided.

  In the present invention, a software program for realizing the above-described embodiment (in the embodiment, a program corresponding to the flowchart shown in the figure) is supplied to the apparatus, and the computer of the apparatus reads the supplied program, Including the case where it is also achieved by executing. Therefore, in order to implement the functional processing of the present invention on a computer, the program itself installed in the computer also implements the present invention. That is, the present invention also includes a speech recognition program for realizing the functional processing of the present invention.

  Thus, when a correct word is input again by voice, a voice recognition device, a voice recognition method, and a voice recognition program that are rich in user convenience without impairing the ease of input inherent in the voice input itself are provided. Can do.

  The configuration described in the above embodiment is merely a specific example and does not limit the technical scope of the present invention. Any configuration can be adopted as long as the effects of the present application are achieved.

  As described above, the speech recognition apparatus according to the present invention has the effect that the user can correct misrecognition without performing a correction operation, and the steps for correcting the misrecognition can be shortened. It is useful as a speech recognition device or the like that corrects the content to the utterance content to be recognized.

Overall block diagram of a speech recognition apparatus in an embodiment of the present invention Sequence diagram for explaining the operation of the speech recognition apparatus according to the embodiment of the present invention An example of the display showing how the speech recognition apparatus according to the embodiment of the present invention displays over time An example of the display showing how the speech recognition apparatus according to the embodiment of the present invention displays over time An example of the display showing how the speech recognition apparatus according to the embodiment of the present invention displays over time An example of the display showing how the speech recognition apparatus according to the embodiment of the present invention displays over time An example of the display showing how the speech recognition apparatus according to the embodiment of the present invention displays over time

Explanation of symbols

100 voice input unit 110 recognition start end unit 200 recognition unit 210 recognition score calculation unit 300 recognition target word storage unit 400 correction timing control unit 500 notification content control unit 600 notification unit D300 display D301 display region D302 display region D303 scale D304 time display region W300 recognition word

Claims (11)

A speech recognition device,
A storage unit for storing word data;
The word data represented by the input speech is compared with the word data stored in the storage unit, a recognition score representing the matching degree at the time of comparison is calculated, and the recognized word is output based on the calculated recognition score A voice recognition unit that
Based on the recognition score calculated by the voice recognition unit, a reception time determination unit that determines a reception time in which re-input by voice can be received;
A speech recognition apparatus comprising: a speech re-recognition unit that re-recognizes a word represented by speech re-input within an acceptance time determined by the reception time determination unit and outputs the re-recognized word.   The speech recognition apparatus according to claim 1, further comprising a reception time notification unit that notifies a user by a display object that changes with a change in the reception time determined by the reception time determination unit.   The voice recognition device according to claim 2, wherein the reception time notification unit stops the change of the display object when the voice is re-input within the reception time determined by the reception time determination unit. .   The voice according to claim 3, wherein the reception time notification unit changes the position of the display object representing the reception time determined by the reception time determination unit on the display area together with the change of the reception time. Recognition device.   The speech recognition apparatus according to claim 3, wherein the reception time notification unit changes a scale amount of a display object representing the reception time determined by the reception time determination unit along with a change in the reception time.   The voice recognition device according to claim 3, wherein the reception time notification unit changes the size of a display object representing the reception time determined by the reception time determination unit with a change in the reception time.   The voice recognition device according to claim 3, wherein the reception time notification unit changes the transparency of the display object representing the reception time determined by the reception time determination unit with a change in the reception time.   The voice re-recognition unit removes words recognized by the voice recognition unit from the storage unit and re-recognizes during the time determined by the acceptance time notification unit. The speech recognition device according to any one of the above.   The speech re-recognition unit re-recognizes a word in which a predetermined word is added to a word recognized by the speech recognition unit during a time determined by the reception time determination unit. The speech recognition device according to any one of the above. A speech recognition method,
The word data represented by the input speech is compared with the word data stored in the storage unit, a recognition score representing the degree of matching at the time of comparison is calculated, and the recognized word is output based on the calculated recognition score. A speech recognition step;
Based on the recognition score calculated in the voice recognition step, a reception time determination step for determining a reception time in which re-input by voice can be received;
A speech recognition method comprising: a speech re-recognition step of re-recognizing a word represented by speech re-input within the reception time determined in the reception time determination step and outputting the re-recognized word. A speech recognition program executed by a computer of a speech recognition device,
In the computer,
The word data represented by the input voice is compared with the word data stored in the storage unit, a recognition score representing the degree of matching at the time of comparison is calculated, and the recognized word is output based on the calculated recognition score. A speech recognition step;
Based on the recognition score calculated in the voice recognition step, a reception time determination step for determining a reception time in which re-input by voice can be received;
A speech recognition program for executing a speech re-recognition step of re-recognizing a word represented by speech re-input within the reception time determined in the reception time determination step and outputting the re-recognized word.

Images