JP2005283797A - Device and method for speech recognition - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech recognition device capable of correcting contents which were misrecognized in speech recognition processing by simple utterance. <P>SOLUTION: The result of the speech recognition processing is displayed on a display together with vocabularies 31 to 33 for correction. When the speech recognition result is corrected from "0482XX8888" to "0482XX2888", for example, the vocabulary 31 for correction is used and "2 for the 7th" is uttered to make a correction. Consequently, the micrecognized part can be corrected by specifying the correction position, so the contents which were uttered first need not be all uttered again and corrections can easily be made. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a speech recognition apparatus and speech recognition method for recognizing speech and converting it into information such as characters.

  In a conventional speech recognition apparatus, when an input speech is erroneously recognized, a technique is known in which a user can read back a misrecognized portion and speak a correction word to make a correction. (See Patent Document 1).

JP 2000-293195 A

  However, in the conventional speech recognition apparatus, for example, when an utterance of “1717” is erroneously recognized as “7777”, even if an utterance for correcting “77 to 17” is made, a plurality of “77” ”, The correction by this method could not be performed, and all parts had to be re-spoken to correct“ 7777 to 1717 ”.

  The voice recognition device and the voice recognition method according to the present invention specify the correction content including at least the correction position specified by the input correction voice when the voice for correcting the voice recognition result is input, and The speech recognition result is corrected.

  According to the speech recognition apparatus and speech recognition method of the present invention, it is possible to specify and correct the position of the misrecognized portion of the speech recognition result, so that the correction can be easily performed.

-First embodiment-
FIG. 1 is a diagram showing a configuration of a voice recognition apparatus according to a first embodiment of the present invention. Hereinafter, an example will be described in which the speech recognition apparatus according to the first embodiment is applied to an in-vehicle hands-free telephone system that is mounted on a vehicle and can make a call by inputting a telephone number by speech. The speech recognition apparatus in the first embodiment includes a speech recognition controller 1, a phone controller 2, a microphone 3, a speech input operation switch 4, a speaker 5, a display 6, and a mobile phone 10.

  The microphone 3 collects the voice spoken by the user and outputs it to the voice recognition controller 1. The voice input operation switch 4 is installed, for example, in the steering of the vehicle, and is operated when the driver performs voice input. The voice recognition controller 1 includes a CPU 1a, a ROM 1b, and a RAM 1c. When the voice input operation switch 4 is operated, the CPU 1a reads an icon indicating a voice standby state from the ROM 1b, displays the icon on the display 6, and outputs a beep sound indicating a utterance start signal from the speaker 5.

  When the user confirms the beep sound output from the speaker 5, the user speaks “dial input”, which is a command for starting dial input of the hands-free telephone, toward the microphone 3. The vocabulary corresponding to “dial input”, which is a command for starting the dial input, is registered in the ROM 1b in advance.

  The CPU 1a of the speech recognition controller 1 compares the speech input via the microphone 3 with the vocabulary in the speech recognition dictionary stored in the ROM 1b, and specifies the input vocabulary. Since such a voice recognition process can use a known method, a detailed description of the voice recognition process is omitted.

  When the CPU 1a recognizes the voice “dial input”, the character “dial input”, which is the recognition result, is displayed on the display 6 and from the speaker 5, the voice “dial input. Is output. The characters to be displayed on the display 6 can be registered in the ROM 1b in advance, and the CPU 1a can read out from the ROM 1b and display them on the display 6. The sound output from the speaker 5 can use, for example, a TTS (Text-To-Speech) function. Also in this case, the message output from the speaker 5 may be stored in the ROM 1b.

  When the voice “Dial input. Please give me the number” is output from the speaker 5, the user speaks the telephone number that the mobile phone 10 wants to call. Here, it is assumed that “0482XX2888” is spoken. The number spoken by the user is input to the speech recognition controller 1 via the microphone 3. Since the voice input information of the telephone number is limited to numbers, the CPU 1a of the voice recognition controller 1 recognizes the input numbers based on the number recognition dictionary stored in the ROM 1b.

  Here, it is assumed that the number recognition result is “0482XX8888”. When the CPU 1a displays “0482XX8888” as a result of the recognition process on the display 6 and counts the number of digits of the recognized number and determines that all digits of the telephone number are input, “0482XX8888 is OK? A message for confirming the recognition result such as “” is output from the speaker 5 by voice.

  FIG. 2 is a diagram showing a message displayed on the display 6 for confirming the recognized telephone number. As shown in FIG. 2, the display 6 displays the recognized telephone number 20, an icon 25 indicating that speech is possible, and response examples 30 to 33 for messages confirming the recognition result. The user can speak while the icon 25 is displayed.

  Response example 30 shows a case where the number recognized by voice recognition is correct. Response example 30, that is, when the user speaks “Yes” and the voice recognition controller 1 correctly recognizes the call, the CPU 1a calls the phone controller 2 so as to call the recognized number. Instructions are given. The phone controller 2 includes a CPU 2a, a ROM 2b, and a RAM 2c. When receiving the call instruction, the CPU 2a of the phone controller 2 issues a command to the mobile phone 10 so as to make a call to the specified number. Thereby, it is possible to make a call hands-free.

  The response examples 31 to 33 show cases where the number recognized by voice is wrong. The response example 31 shows a case where a wrong place is designated by a number and a correct number is input. In the above-described example, although “0482XX2888” is spoken, it is recognized as “0482XX8888”, so the seventh (seventh digit) number from the head is erroneously recognized. Therefore, by speaking “7th to 2”, the 7th number can be corrected to “2”.

  The response example 32 indicates a case where a wrong part number and a wrongly recognized number are designated and a correct number is input. In the example described above, the seventh number “8” can be corrected to “2” by saying “7th 8 is 2”. The response example 33 indicates a case where a wrong number is designated and a correct number is input. For example, by speaking “8888 to 2888”, a misrecognized portion can be corrected. However, since there are a plurality of numbers “8” among the recognized numbers, “8 is 2” cannot be corrected by speaking.

  The vocabulary corresponding to the response examples 30 to 33 is registered in advance in the ROM 1b. The CPU 1a compares the input voice with the vocabulary corresponding to the response examples 30 to 33 registered in the ROM 1b to determine whether the vocabulary corresponding to the response example 30 that affirms the recognition result has been input or for correction. It is determined whether the vocabulary corresponding to the response examples 31 to 33 is input.

  FIG. 3 is a flowchart showing the contents of processing performed by the speech recognition apparatus according to the first embodiment. The process starting from step S10 is performed by the CPU 1a of the speech recognition controller 1. In step S10, it is determined whether or not the voice input operation switch has been operated. If a signal indicating that the voice input operation switch 4 has been pressed is input, the process proceeds to step S20. If it is determined that the voice input operation switch 4 has not been pressed, the process waits in step S10.

  In step S20, an utterable icon (speech standby icon) is read from the ROM 1b and displayed on the display 6, and a beep sound is output from the speaker 5. In step S30 following step S20, it is determined whether or not voice input by the user has been performed. If it is determined that voice input has been performed via the microphone 3, the process proceeds to step S40. If it is determined that voice input has not been performed, the process waits in step S30.

  In step S40, speech recognition processing is performed on the words that are input by speech, and the process proceeds to step S50. In step S50, it is determined whether or not the speech-recognized word is “dial input” as a result of the speech recognition processing in step S40. If it is determined that the speech-recognized word is “dial input”, the process proceeds to step S60. If it is determined that the word is other than “dial input”, the process returns to step S10.

  In step S60, the character “dial input”, which is the recognition result, is displayed on the display 6 and the speaker 5 outputs a voice “dial input. In step S70 following step S60, it is determined whether voice input has been performed via the microphone 3. If it is determined that voice input has been performed, the process proceeds to step S80. If it is determined that voice input has not been performed, the process waits in step S70.

  In step S80, voice recognition processing is executed for the input voice, and the process proceeds to step S90. In step S90, the speaker 5 outputs a voice asking the user whether the result of the voice recognition performed in step S80 is correct. For example, when the phone number recognized as voice is “0482XX8888”, a voice such as “Are you sure it is 0482XX8888” is output from the speaker 5. At the same time, the phone number that is the recognition result and the wording for correction are displayed on the display 6 (see FIG. 2).

  In step S100, it is determined whether or not the user has responded to the voice asking whether the voice recognition result output in step S90 is correct. If it is determined that the user's voice is input via the microphone 3, the process proceeds to step S110. If it is determined that no voice is input, the process waits in step S100. In step S110, voice recognition processing is performed on the input voice, and the process proceeds to step S120.

  In step S120, it is determined whether or not the input vocabulary is a correction vocabulary as a result of the speech recognition process in step S110. As described above, this determination is made based on whether or not the vocabulary corresponding to the correction response examples 31 to 33 registered in the ROM 1b is input. If it is determined that a corrected vocabulary such as “7th is 2” is input, the process proceeds to step S130. If it is determined that the input vocabulary is not a vocabulary for correction, the process proceeds to step S140.

  In step S130, the telephone number recognized in step S80 is corrected based on the input correction vocabulary. When the correction process is executed, the process returns to step S90, and a voice asking the user whether or not the corrected telephone number is correct is output via the speaker 5. Also in this case, the corrected telephone number and the wording for correction are displayed on the display 6.

  In step S140, it is determined whether or not the input vocabulary is a vocabulary corresponding to the response example 30 that affirms the recognition result. If it is determined that the vocabulary that affirms the recognition result, that is, the vocabulary “Yes” is input, the process proceeds to step S150. On the other hand, if it is determined that the input vocabulary is not a vocabulary that affirms the recognition result, the process returns to step S100.

  In step S150, a call instruction is issued to the phone controller 2 so as to make a call to the telephone number recognized when the vocabulary that affirms the recognition result is input. The CPU 2a of the phone controller 2 that has received the call command causes the mobile phone 10 to make a call to the phone number that has received the command.

  According to the speech recognition apparatus in the first embodiment, when the speech recognition result for the speech input by the user is incorrect, the position to be corrected can be designated and corrected. Can do. That is, since it is not necessary to utter again all the contents including the correctly recognized part in order to correct the speech recognition result, the convenience at the time of correction is enhanced.

  In the conventional speech recognition apparatus, when the same number is included in the number string, it is impossible to specify and correct only the number. For example, in order to correct “8888” to “2888”, it was not possible to correct it by uttering “8 to 2”. However, according to the speech recognition apparatus in the first embodiment, “1st Can be corrected by saying “2” or “1st 8 is 2”.

  Further, since the vocabulary for correcting the speech recognition result (speech input example for correction) is displayed on the display 6, the user does not need to store the vocabulary for correction and is displayed. Corrections can be made easily using vocabulary.

-Second Embodiment-
In the voice recognition apparatus according to the first embodiment, an example has been described in which a user utters a telephone number to be input at a time. In the speech recognition apparatus according to the second embodiment, an example will be described in which a user does not utter a telephone number at a time, but utters by dividing it by several digits. The configuration of the speech recognition apparatus in the second embodiment is the same as the configuration of the speech recognition apparatus in the first embodiment shown in FIG.

  For example, assume that the user speaks “046”, which is the first three digits of the telephone number. The CPU 1a of the speech recognition controller 1 performs speech recognition processing on the input speech, displays the recognition result on the display 6, and, as with the speech recognition apparatus in the first embodiment, corrects vocabulary for correction. indicate. At this time, the CPU 1a counts the number of digits of the voice-recognized numbers, and when determining that all the digits of the telephone number have not been input, outputs a message prompting the voice input of the subsequent numbers. For example, if the recognized number is “046”, a message “Please continue with 046” is output from the speaker 5.

  Next, assume that the user utters “270”, which is the next three-digit number. The CPU 1a of the speech recognition controller 1 performs speech recognition processing on the input speech, displays the recognition result on the display 6, and displays the vocabulary for correction.

  FIG. 4 is a diagram showing the telephone number 20, the utterable icon 25, and correction response examples 31 to 33 and 40 recognized so far, which are displayed on the display 6. The correction response examples 31 to 33 are the same as the correction response examples 31 to 33 shown in FIG. However, since all the digits of the telephone number are not input, the response example 30 (see FIG. 2) that confirms the voice recognition result is not displayed.

  The correction response example 40 is a vocabulary for additionally inputting numbers when the number of digits in the number string that is the result of the speech recognition process is smaller than the number string input by voice. Generally, when a certain number string is input by voice, the first number has a small utterance power and may not be recognized by the voice recognition process. For example, even when “270” is input as a voice, the first number “2” may not be recognized and may be erroneously recognized as “70”.

  In such a case, if the user speaks “add 2” using the vocabulary shown in the response example 40, “2” is added before “70” which is the result of the immediately preceding speech recognition process. Corrections can be made. It should be noted that correction can be performed by speaking “70” like “270” using the vocabulary shown in the response example 33.

  FIG. 5 is a flowchart showing the processing contents performed by the speech recognition apparatus according to the second embodiment, which is performed by the CPU 1a of the speech recognition controller 1. The same processes as those performed by the speech recognition apparatus in the first embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.

  The processing from step S10 to step S80 is the same as the processing from step S10 to step S80 in the flowchart shown in FIG. In step S200 following step S80, a number that is the result of the speech recognition process performed in step S80 is displayed on the display 6. For example, if the recognized number is three digits, a three-digit number is displayed on the display 6. At this time, as described with reference to FIG. 4, the response examples 31 to 33 and 40 for correction are also displayed at the same time.

  In step S210 following step S200, it is determined whether or not all digits of the telephone number have been input based on the number sequence that has been voice-recognized. If it is determined that all digits of the telephone number have been input, the process proceeds to step S90. If it is determined that all digits have not been input, the process proceeds to step S220. In step S <b> 220, a voice prompting voice input of numbers following the telephone number is output from the speaker 5.

  In step S230 following step S220, it is determined whether or not a voice input of a number following the telephone number has been performed. If it is determined that the user's voice is input via the microphone 3, the process proceeds to step S240. If it is determined that no voice is input, the process waits in step S230. In step S240, voice recognition processing is performed on the input voice, and the process proceeds to step S250.

  In step S250, as a result of the speech recognition process performed in step S240, it is determined whether or not the word input by speech in step S230 is a number. If it is determined that the speech input word is a number, the process returns to step S200, and the number that is the result of the speech recognition process is displayed on the display 6. On the other hand, if it is determined that the speech input word is not a number, the process proceeds to step S120.

  In step S120, it is determined whether or not the word input by voice in step S230 is a vocabulary corresponding to the response examples 31 to 33 and 40 for correction. If it is determined that the speech input word is a vocabulary corresponding to the correction response examples 31 to 33, 40, the process proceeds to step S130. If it is determined that the word is not a vocabulary corresponding to the correction response example, the process returns to step S230. . In step S130, telephone number correction processing is executed based on the input correction vocabulary. When the correction process is executed, the process returns to step S200, and the corrected telephone number is displayed on the display 6.

  In step S90, which proceeds after it is determined that all digits of the telephone number have been input, a voice asking the user whether or not the voice recognition result is correct is output from the speaker 5, such as "Are you sure you want to use 046270XXXX?" To do. When the sound for confirming the recognition result is output, the process proceeds to step S100. In step S100, it is determined whether or not there is a user response to the voice for confirming the recognition result. If it is determined that the user's voice has been input, the process proceeds to step S110 to perform voice recognition processing.

  In step S140 following step S110, it is determined whether or not a vocabulary that affirms the recognition result has been input. If it is determined that the vocabulary 30 that affirms the recognition result, that is, the vocabulary “Yes” is input, the process proceeds to step S150. On the other hand, if it is determined that the input vocabulary is not the vocabulary 30 that affirms the recognition result, the process proceeds to step S120.

  In step S150, a call command is issued to the phone controller 2 so as to call the telephone number when the vocabulary that affirms the recognition result is input. The CPU 2a of the phone controller 2 that has received the call command causes the mobile phone 10 to make a call to the phone number that has received the command.

  According to the speech recognition apparatus in the second embodiment, as in the speech recognition apparatus in the first embodiment, it is possible to specify a correction position and correct the erroneously recognized speech recognition result. As a result, correction can be easily performed. Further, since the vocabulary for correcting the speech recognition result is displayed on the display 6, the user does not need to memorize the vocabulary for correction and can easily make corrections using the displayed vocabulary. It can be carried out.

  According to the speech recognition apparatus in the second exemplary embodiment, when a response example 40 for adding a number is determined in advance and a vocabulary corresponding to the response example 40 is input, a result that has already been recognized. On the other hand, since a correction for adding a number is performed, an unrecognized number can be easily added. For example, if a number with a large number of digits, such as a phone number, is input by voice and the first number is not recognized, it is not necessary to input all the digits again, so correction is easy. Can do. As described above, since the first number of utterances has low utterance power, it is assumed that there are many cases where the first number is not recognized. Therefore, if the correction response example 40 for adding the first number of the utterance is determined in advance, an efficient correction method can be realized.

  The present invention is not limited to the embodiments described above. For example, in the speech recognition apparatus according to the first and second embodiments, the example in which the telephone number is input by voice has been described. However, the input voice is not limited to numerals, and may be alphabets, katakana, hiragana, or kanji. Or other characters. Even in this case, for example, when “Kanagawa” is spoken, if it is erroneously recognized as “Kanagawa”, it can be corrected by specifying a correction position such as “Kana” as the first one. .

  The method of specifying the correction position for the speech recognition result is not limited to the vocabulary “Xth”, but may use “X” at the “○” digit, or a vocabulary such as “×” at the “○” character. . Further, the method of specifying the correction part is not limited to the number from the top, but may be a number counted from the last number (character). For example, as a method of correcting “0482XX8888” to “0482XX2888”, the method of correcting “7th to 2” has been described. However, “4th from the right is 2” may be corrected by speaking.

  The display 6 shows response examples 30 to 33 for correcting the speech recognition result (see FIG. 2). For example, a specific correction example such as “second is 1” is shown. Also good. For example, in the case where “1” and “7” are likely to be recognized incorrectly, a specific correction example such as “2 to 1” is shown for the second number recognized as “7”. Thus, correction can be performed more easily.

  After all the digits of the telephone number have been input and a sound confirming the recognition result such as “Are you sure? 0482XX8888” is output from the speaker 5, the user denies the recognition result such as “No”. When audio is input, audio such as “Please correct” may be output from the speaker 5.

  The correspondence between the constituent elements of the claims and the constituent elements of the first and second embodiments is as follows. That is, the microphone 3 is a voice input means, the voice recognition controller 1 is a voice recognition means, a corrected voice input determination means, a correction content specifying means, a correction means and an additional corrected voice determination means, a speaker 5 and a display 6 are output means, The display 6 constitutes display means. In addition, unless the characteristic function of this invention is impaired, each component is not limited to the said structure.

The figure which shows the structure in 1st Embodiment of the speech recognition apparatus by this invention. Figure showing a message on the display confirming the recognized phone number The flowchart which shows the processing content performed by the speech recognition apparatus in 1st Embodiment. Figure showing the phone number and correction vocabulary recognized so far on the display The flowchart which shows the processing content performed by the speech recognition apparatus in 2nd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Voice recognition controller 2 ... Phone controller 3 ... Microphone 4 ... Voice input operation switch 5 ... Speaker 6 ... Display 10 ... Mobile phone

Claims (5)

Voice input means for inputting voice;
Voice recognition means for recognizing the voice input by the voice input means;
Output means for outputting a voice recognition result by the voice recognition means;
Corrected voice input determination means for determining whether or not a voice for correcting the voice recognition result output by the output means (hereinafter referred to as corrected voice) is input;
When it is determined that the corrected voice is input by the corrected voice input determining means, the correction content specifying means for specifying the correction content including at least the correction position designated by the corrected voice;
A speech recognition apparatus comprising: correction means for correcting the speech recognition result at the specified correction position based on the correction content specified by the correction content specifying means. The speech recognition apparatus according to claim 1,
The speech recognition apparatus according to claim 1, wherein the correction position of the speech recognition result is designated by the order of arrangement order of characters or the like constituting the speech recognition result. The speech recognition apparatus according to claim 1 or 2,
Further comprising an additional correction instruction determination means for determining whether or not a correction instruction for adding content to the voice recognition result is included in the corrected voice;
When the correction means determines that the correction instruction for adding content is included by the additional correction instruction determination means, the correction means adds the correction content specified by the correction content specification means to the speech recognition result. A speech recognition apparatus characterized by The speech recognition apparatus according to claim 1 or 2,
The speech recognition apparatus further comprising display means for displaying at least one speech input example for correcting the speech recognition result. Perform voice recognition processing on the input voice,
Outputting the result of the speech recognition process;
It is determined whether or not a voice for correcting the output voice recognition result (hereinafter referred to as a corrected voice) is input,
When it is determined that the corrected voice is input, the correction content including at least the correction position designated by the corrected voice is specified,
A speech recognition method, wherein a speech recognition result at a specified correction position is corrected based on the specified correction content.

Images