JP2003015688A - Voice recognition method and apparatus - Google Patents

Abstract

(57) [Summary] [Problem] In a speech recognition method with a talkback function,
The expression content of the recognition result is "display xxx", and the voice input operation has to be repeated when erroneous recognition occurs. SOLUTION: A similarity between voice data of an input word and a voice pattern in a word dictionary is calculated, five recognition result candidates are acquired in descending order of similarity, and talkback is performed one by one in descending order of similarity. If the similarity is higher than the criterion, a positive expression will be displayed as "○○○". If the similarity is lower than the criterion, the expression will be "○○○？" And ask the user to confirm by voice whether the recognition result is correct or incorrect,
If it is correct, the recognition result is determined. If it is incorrect, the recognition result of the next candidate is displayed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition method and device having a talkback function for confirming an input operation, and an on-vehicle navigation device.

[0002]

2. Description of the Related Art Conventionally, in this type of vehicle-mounted navigation device, setting of a destination, setting of a route, search of facilities, etc.
In addition to the touch panel provided on the front surface of the display or remote control operation, voice recognition technology can be used to easily perform voice input. For example, by simply pressing the voice input button and then voice inputting “go to XX”, the recommended route with the destination set as the destination is displayed on the map on the display. In addition, it is possible to search for a desired facility or residence by voice inputting the facility name, address, telephone number, zip code, etc. in several stages. The voice input in such an in-vehicle navigation device can be greatly contributed to safe driving because the user can input while watching the front for driving the vehicle and without releasing the hand from the steering wheel. In addition, the talkback function that outputs voice in response to the voice input to the device allows you to check whether the content input by the user matches the content determined by the device, which is very convenient. is there.

[0003]

[Problems to be Solved by the Invention] However, in the above-mentioned conventional voice recognition method with a talkback function, for example, when a voice input is made for "Hachioji Station", the result of recognition is "display Hachioji Station" by the talkback. Is output by voice, but the expression content in that case is fixed. The user may input “Hachioji station” by voice, and as a result of voice recognition, the Hachioji station may be correctly recognized, but correct voice recognition is not always possible. The speaker's pronunciation may be unclear or external noise may mix in, resulting in erroneous recognition. Therefore, normally, re-input is possible up to three times. Therefore, if the voice is entered as "Hachioji Station" but it is mistakenly recognized as "Oji Station" and the talk back "Display Prince Station" is made, the user operates the remote control again. The voice input must be redone, which not only increases the burden but also makes the device unsatisfactory.

The present invention solves such a conventional problem, improves the reliability of the device, is easy to operate, and is a convenient voice recognition method and device, and in-vehicle navigation using the same. The purpose is to provide a device.

[0005]

In order to achieve the above object, the speech recognition method of the present invention calculates the degree of similarity between the voice data of an input word and the voice pattern in the word dictionary to obtain a high degree of similarity. A plurality of voice patterns are sequentially used as word recognition candidates, and when the input operation is confirmed by outputting the word recognition candidates by voice, the expression content of the voice output depending on whether the similarity is higher or lower than a reference is displayed. It is characterized by changing. Similarity calculations are usually from 0 to 9
999 calculation results are obtained, and the higher the score, the higher the similarity. Therefore, for example, it is empirically known that the possibility of erroneous recognition increases when the number of points is 3000 points or less and the possibility of erroneous recognition decreases when the number of points exceeds 3000. By changing the content expressed by the talkback based on the score, the user can understand that the device makes a proper judgment and can improve the reliability of the device.

Further, in the voice recognition method of the present invention, when the similarity is higher than a reference, the expression content of the output voice is affirmative, and when the similarity is lower than the reference, the output is performed. It is characterized in that the expression content of the voice that is expressed is a question form. For example, if the similarity is high, "○
"○○ is displayed", and when the similarity is low, by asking "Is it ○○○?"
The user understands that the device is making the right decisions,
The reliability of the device can be further increased.

Further, in the voice recognition method of the present invention, the user is asked to confirm whether the output word is “correct” or “not” by the output voice, and when it is “correct”, the output word is confirmed.
In the case of "no", it is characterized in that the next candidate word is output by voice, and even if an erroneous recognition occurs, it is not necessary to input the voice again as in the conventional case. Since the next candidate is displayed only by uttering, the burden on the user is small and the usability can be improved.

Further, the voice recognition apparatus of the present invention calculates the similarity between the voice data of the input word and the voice pattern in the word dictionary, and outputs a plurality of voice patterns as word recognition candidates in descending order of similarity. A voice recognition means for confirming the input operation, and a control means for changing the expression content by the output voice depending on whether the similarity is higher or lower than a reference when confirming the output voice. For example, if the similarity is high, "○○○
Is displayed. ”And when the degree of similarity is low,“ ○○
By asking the question "Is it o?", The user can understand that the device is making a proper judgment and can further enhance the reliability of the device.

Further, in the voice recognition device of the present invention, the control means asks the user for confirmation of "correct" or "not" of the output word by the output voice, and when it is "correct", the output word is correct. Is determined, and in the case of “no”, the next candidate word is output by voice, and even if an erroneous recognition occurs, it is not necessary to input voice again as in the conventional case. Since the next candidate is displayed only by uttering “No”, the burden on the user is reduced and usability can be improved.

Further, the present invention is a vehicle-mounted navigation device equipped with the above-mentioned voice recognition device, and the convenience and operability of the navigation device can be further improved.

[0011]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of a vehicle-mounted navigation device equipped with a voice recognition device according to an embodiment of the present invention. In FIG. 1, the orientation sensor 1 is
A 3D gyro is used to detect the heading of the vehicle.
The vehicle speed sensor 2 is used in an electronic control unit of a vehicle equipped with this device, and generates a vehicle speed pulse according to the number of rotations of wheels. Various sensors 3 are reverse switches,
It is a parking switch, a light switch, etc., and detects the running state of the vehicle. The sensor signal processing unit 4 calculates the traveling direction of the vehicle based on the signal from the azimuth sensor 1, the traveling distance from the vehicle speed signal from the vehicle speed sensor 2, and further the vehicle based on the signals from the various sensors 3. Detects the traveling state of the vehicle and generates a signal required for control. DVD-
The ROM 5 stores map data, voice data, voice recognition dictionary data, and the like. DVD-ROM drive 6
Is for reading map data, voice data, voice recognition dictionary data, etc. from the DVD-ROM 5. The liquid crystal display 7 displays a map and the current vehicle position, direction, operation menu, and the like, and may have an operation unit such as a touch panel on the front surface thereof. The GPS receiver 8 obtains the current position (latitude / longitude) of the vehicle by receiving and calculating radio waves transmitted from a plurality of satellites. The GPS antenna 9 is an antenna for receiving GPS radio waves. The DVD-ROM drive 6, the liquid crystal display 7, the GPS receiver 8 and the like are arranged on the dashboard of the vehicle and are connected to the communication interface 12 of the device body 11 through the in-vehicle LAN 10. The device main body 11 is installed in a trunk room of a vehicle or a center console in the vehicle.

The microphone 13 is arranged in the vicinity of the driver in the vehicle and is used for inputting a vocal phrase from the user, and the speaker 14 is used for a search result, a voice recognition result, intersection guidance on a traveling route, branch guidance, It is used for voice guidance such as toll booth guidance and exit guidance, and for confirming the operation contents with the remote control by voice. The voice recognition device 15 performs voice recognition of the phrase input from the microphone 13. The storage unit 16 includes a ROM that stores a program, a RAM that temporarily stores work data, a VRAM that stores image data, and the like. The image processor 17 performs display image generation processing based on menu data, map data, current position data of the vehicle, building data, and the like. Display control unit 18
In the normal setting mode and the voice recognition mode, the necessary image data is read from the image processor 17 and passed to the CPU 20. Voice processor 19
Converts the voice recognition result into a voice signal, and outputs a voice signal indicating the search result, voice guidance on the traveling route, and operation content of the remote controller to the speaker 14. A CPU (Central Processing Unit) 20 controls the entire apparatus, and executes programs for performing necessary control in the normal setting mode and the voice recognition mode, respectively. The remote controller 21 has an operation button for switching between the normal setting mode and the voice recognition mode and other operation buttons, and communicates with the remote controller light receiving unit 22 using infrared rays. Although the remote controller light receiving unit 22 is provided on the front surface of the liquid crystal display 7, it may be provided at another position, and sends an operation signal received from the remote controller 21 from the in-vehicle LAN 10 to the CPU 20 via the communication interface 12.

Next, regarding the operation of the present embodiment, the basic operation of the navigation device will be described first. In FIG. 1, when the engine of the vehicle is started,
The navigation device is powered on, a menu screen is displayed on the liquid crystal display 7, and the CPU 20 activates the current position detection program. When the vehicle starts traveling, the position information from the GPS receiver 8 and the direction sensor 1
The accurate current position of the vehicle is calculated based on the data obtained by processing the signal from the vehicle speed sensor 2 by the sensor signal processing unit 4. Based on this vehicle position information, the CPU 20
Corresponding map data is read from the DVD-ROM 5 through the VD-ROM drive 6, converted into image data by the image processor 17 and temporarily stored in the VRAM of the storage unit 16, and then converted into a color signal for communication interface 1.
Displayed together with the vehicle position on the screen of the liquid crystal display 7 through 2. When an address name such as a destination is input through the microphone 13, the voice recognition function of the voice recognition device 15 recognizes the address name and the destination is set. When the destination is set, the CPU 20 activates the route search program, calculates the optimum guide route from the current position of the vehicle to the set destination, and displays it on the map of the liquid crystal display 7 in an overlapping manner. . When the user advances the vehicle along the guide route displayed on the liquid crystal display 7, the CPU 20
Will sequentially update the vehicle position mark on the liquid crystal display 7 based on the current position information and the road network data on the map data. When the vehicle approaches a branch point or the like in the guide route, the voice guide added to the map data is output from the speaker 14.

Next, the voice recognition device 15 in the above embodiment will be described. FIG. 2 shows the configuration of the voice recognition device 15. The voice analysis unit 23 frequency-analyzes the voice input from the microphone 13 and outputs it. Word dictionary section 2
In 4, a standard pattern of word voice is stored. The voice collation unit 25 calculates the similarity between the input voice data output from the voice analysis unit 23 and the standard pattern of the word output from the word dictionary unit 24, and the recognition results are calculated in descending order of similarity between the two. Output as. The control unit 26 is a microcomputer that controls the entire voice recognition device 15, and the DV is transmitted through the CPU 20 of the vehicle-mounted navigation device.
The word dictionary data read from the D-ROM 5 is stored in the word dictionary unit 24, read at the time of matching in the voice matching unit 25, and the recognition result output from the voice matching unit 25 is converted into synthetic speech via the voice processor 19. Speaker 1
The process of outputting from 4 is performed.

Next, the control in the control unit 26 will be described with reference to the flow chart of FIG. When a voice is input from the microphone 13 (step S1), the voice is frequency-analyzed for each word by the voice analysis unit 23 and output as a time series of a short-time spectrum (LPC cepstrum coefficient sequence) (step S2). The word dictionary unit 24 stores a standard pattern of about 3000 words consisting of a phoneme symbol series (LPC cepstrum coefficient sequence). The voice matching unit 25
The similarity between the time series of the short-term spectrum of the word output from the speech analysis unit 23 and the phoneme symbol sequence of the word output from the word dictionary unit 24 is calculated, and the score is high from zero to 9999 calculation results. That is, five phoneme symbol sequences are output as recognition results in descending order of similarity (step S
3), the top five lists are created by arranging them in descending order of similarity score (step S4). Whenever the recognition result is output from the voice collation unit 25, the control unit 26,
It is checked whether or not the similarity score exceeds 3000 points (step S5), and the recognition result output exceeding 3000 points is classified into the talkback expression pattern A (step S6) and exceeds 3000 points. The recognition result output that does not exist is classified into the talkback expression pattern B (step S7). As shown in FIG. 4, the expression pattern A is composed of an affirmative sentence "Display XX.", And the expression pattern B is "XX? Is correct? Yes, if wrong," It consists of an indirect question and a confirmation that say "No". After classifying in this way, one by one is output by talkback in order from the top (step S8). For example, when "Tokyo Dome" is input by voice, the voice recognition result is as shown in FIG. 5, "Tokyodo" is 2900 points, "Tokyo Dawn" is 2300 points,
1800 points for "Tokyo Dome" and 150 points for "Tokyo Doll"
If there are 0 points and 1000 points in "Tokyo", first of all, the talkback will be output as "Display Tokyo-do. If correct, please say" Yes ", if wrong, please say" No "" ( Step S9). On the other hand, when the user utters "No", the voice recognition device 15 recognizes the voice and the control unit 26 extracts the next voice recognition result candidate from the list corresponding to "No", Dawn? If yes, please say "yes", and if not, please say "no". " Also, in response to this, when the user utters "No", the control unit 26 extracts the next voice recognition result candidate and asks, "Is this the Tokyo Dome? "No, please say", and the talkback is output as voice. On the other hand, when the user utters "Yes", the control unit 26 determines "Tokyo Dome" as the voice recognition result (step S10). Although the voice recognition process is finished with this, as a navigation function thereafter, three selection items of “destination”, “route setting”, and “point registration” are displayed on the liquid crystal display 7, and “destination setting or route setting” is displayed. The voice guidance "Whether you want to register or point registration" is output, and the user selects one of them to perform the setting registration for the searched facility.

As described above, according to the present embodiment, the similarity between the voice data of the input word and the voice pattern in the word dictionary is calculated, and five recognition result candidates are listed in descending order of similarity. It is held and output one by one in descending order of similarity, and if the similarity score exceeds 3000 points, it is expressed as an affirmative expression, and if the similarity score is less than 3000 points, it is used as an indirect question expression, and the recognition result Asks the user to confirm whether or not is correct or incorrect, so that the user can understand that the device is making a correct decision, and it is not necessary to repeat the voice input operation many times as in the past, so that the device can be trusted. The operability and operability can be improved.

Note that the talkback expression pattern in the above embodiment is an example, and various expressions other than these are possible. Further, the setting of the similarity score when changing the expression content of the talkback is variable, and the number of expression patterns of the talkback and the number of words to be included in the voice recognition result list can be set arbitrarily.

[0018]

As described above, the speech recognition method and apparatus of the present invention calculates the degree of similarity between the speech data of an input word and the speech pattern in the word dictionary, and a plurality of speeches are arranged in descending order of similarity. Output patterns as word recognition candidates,
When confirming the input operation by the output voice, the content of the output voice is changed depending on whether the similarity is higher or lower than the reference, so that the user can properly recognize the voice according to the similarity of the device. Therefore, it is possible to improve the reliability of the device.

Further, the user is asked to confirm whether the output word is "correct" or "not" by the output voice. If the word is "correct", the output word is confirmed, and if it is "no", the next candidate is selected. Since words are output by voice, even if erroneous recognition occurs, there is no need to repeat the voice input operation as in the past, for example, just saying "No" will display the next candidate,
The burden on the user is small and usability can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a vehicle-mounted navigation device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a voice recognition device according to an embodiment of the present invention.

FIG. 3 is a flowchart showing processing of the voice recognition device in the embodiment of the present invention.

FIG. 4 is a table diagram showing an expression pattern in talkback according to the embodiment of the present invention.

FIG. 5 is a table diagram showing a list of voice recognition results based on similarity score according to the embodiment of the present invention.

[Explanation of symbols]

1 Direction sensor 2 vehicle speed sensor 3 Various sensors 4 Sensor signal processing unit 5 DVD-ROM 6 DVD-ROM drive 7 LCD display 8 GPS receiver 9 GPS antenna 10 Car LAN 11 Device body 12 Communication interface 13 microphone 14 speakers 15 Speech recognition device 16 Memory 17 Image Processor 18 Display control unit 19 voice processor 20 CPU 21 remote control 22 Remote receiver 23 Speech analysis section 24 word dictionary 25 Voice verification unit 26 Control unit

   ─────────────────────────────────────────────────── ─── Continued front page    F term (reference) 2F029 AA02 AB01 AB07 AB09 AB13                       AC02 AC08 AC14 AC18                 5D015 KK01 LL06                 5H180 AA01 BB13 EE01 FF04 FF05                       FF22 FF25 FF27 FF33

Claims (6)

[Claims] 1. A degree of similarity between voice data of an input word and a voice pattern in a word dictionary is calculated, a plurality of voice patterns are set as word recognition candidates in descending order of similarity, and the word recognition candidates are output by voice. Thus, when confirming an input operation, a voice recognition method is characterized in that the expression content of the output voice is changed depending on whether the similarity is higher or lower than a reference. 2. When the similarity is higher than a reference, the expression content of the output voice is affirmative, and when the similarity is lower than the reference, the expression content of the output voice is doubtful. The speech recognition method according to claim 1, wherein the speech recognition method has a shape. 3. The output voice prompts the user to confirm whether the output word is “correct” or “not”. When the word is “correct”, the output word is confirmed, and when it is “not”, the next candidate is selected. 3. The word of 1 is output as a voice.
The voice recognition method described. 4. The input operation is confirmed by calculating the similarity between the voice data of the input word and the voice pattern in the word dictionary, and outputting a plurality of voice patterns as word recognition candidates in descending order of similarity. A voice recognition device comprising: a voice recognition unit; and a control unit that changes the expression content of the output voice depending on whether the similarity is higher or lower than a reference when performing confirmation by the output voice. 5. The control means requests the user to confirm whether the output word is “correct” or “not” by the output voice,
5. The voice recognition device according to claim 4, wherein the output word is determined when the result is “positive”, and the next candidate word is output by voice when the result is “no”. 6. An in-vehicle navigation device equipped with the voice recognition device according to claim 4.