JP2005227369A - Voice recognition device, and method, and vehicle mounted navigation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition device with which voice recognition by dialect can be performed. <P>SOLUTION: The voice recognition device which collates the result obtained by subjecting the voice of an inputted word to a frequency analysis with a word dictionary formed by using a plurality of recognition templates and recognizes the voice by extracting the word of a high degree of similarity is provided. A DVD-ROM 5 is provided with a standard template and a plurality of area templates. A CPU 20 switches the recognition template to an arbitrary template by selection of a user. The user is able to carry out voice recognition by selecting the area template of the area dialect suitable for the user and the voice recognition is made possible even by using the area language. Convenience and operability can be improved. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a speech recognition device and method for recognizing speech input by a user, and an in-vehicle navigation device including the speech recognition device.

2. Description of the Related Art Conventionally, in order to further improve convenience, a technique for performing operations such as a car air conditioner, a car audio, and a car navigation mounted on a vehicle using a voice recognition device is known (for example, see Patent Document 1). . What is important in the speech recognition apparatus is to prevent misrecognition and increase the recognition rate. As one measure to prevent the occurrence of misrecognition, three recognition templates for men, women and men and women are prepared as recognition templates for matching the short-time spectrum of the input recognition target word with the standard pattern. Thus, a technique is known in which a recognition template suitable for each gender is used to increase the recognition rate in speech recognition as compared to the case where a single recognition template is used.
JP 2002-169484 A

  However, even if three recognition templates are prepared for men, women, and mixed in this way, they are all targeted for standard words, and accents, intonations, phrases are different, and the pronunciation is pronounced by the user There is a problem that the recognition rate does not improve so much.

  The present invention solves such a conventional problem, and a speech recognition apparatus and method capable of improving the recognition rate even when a user with a different accent, intonation, or phrase, or with a rough sound, can improve the recognition rate. It is an object of the present invention to provide an in-vehicle navigation device equipped with a voice recognition device.

  The speech recognition apparatus of the present invention compares the result of frequency analysis of the speech of the input word with a word dictionary created using a plurality of recognition templates, extracts speech with high similarity, and recognizes speech A recognition apparatus comprising a standard template and a plurality of regional templates as the recognition templates, and a recognition control means capable of arbitrarily switching the recognition templates.

  With this configuration, if the user is aware that his / her pronunciation is slightly different from the standard language, voice input can be facilitated by selecting a regional template in a regional language suitable for him / her and performing speech recognition. In addition, the voice recognition rate can be increased, and the operability can be improved by reducing the number of inputs in voice recognition.

  Moreover, the speech recognition apparatus of the present invention is characterized in that the recognition template includes a plurality of templates for men, women, and a mixture, respectively.

  With this configuration, the user can select the male or female type according to his / her gender or voice quality, or if the male and female are together, the recognition template for mixing can be selected to further increase the recognition rate. be able to.

  The speech recognition apparatus of the present invention is characterized in that the recognition control unit switches the recognition template based on a selection item designated by a user.

  With this configuration, the recognition control unit selects a recognition template according to the selected selection item by selecting in advance from the selection items which recognition template is used for voice recognition. be able to.

  In the speech recognition apparatus of the present invention, the selection item is regional data selected from a regional code displayed on a display unit.

  With this configuration, the user can select a regional template for the region by selecting one of the regional codes displayed on the screen of the display means, for example, a regional code assigned to each prefecture in Japan. it can.

  In the voice recognition device according to the present invention, the selection item is a license read by a card reading means, or permanent data or address data from an IC card.

  With this configuration, when the user inserts an IC card type license or a credit-type or prepaid type IC card into the card reading means, the card reading means has a permanent address or current address written on the license, or an IC card. By reading the current address described in the above and converting it into the code data of the region, the region template of the region can be selected.

  In the speech recognition apparatus of the present invention, the selection item is current position data from a current position detecting unit that detects a current position.

  With this configuration, the user can automatically select a region template corresponding to the current position as the user moves by selecting current position data from the current position detecting unit that detects the current position as a selection item. it can.

  In the voice recognition device of the present invention, the selection item is home position data registered by a point registration unit.

  With this configuration, the user can select an area template corresponding to the home position by selecting the home position data registered by the point registration unit as a selection item.

  The speech recognition apparatus according to the present invention is characterized in that the recognition control means has a learning function for storing a recognition template having a high recognition rate used last time and using it as a next recognition template.

  With this configuration, the user selects which selection item, and the selected regional template stores which regional template is for men, women, or mixed, and the region with the highest recognition rate By storing the template, the next time the voice recognition device is activated, the same regional template as used last time can be used, and the recognition efficiency and operability can be improved.

  Also, the speech recognition method of the present invention recognizes speech by comparing the result of frequency analysis of the speech of the input word with a word dictionary created using a plurality of recognition templates and extracting words with high similarity. A method is characterized in that a standard template and a plurality of regional templates are provided as the recognition templates, and voice recognition is performed by arbitrarily switching the recognition templates.

  With this configuration, if the user is aware that his / her pronunciation is a little different from the standard language, speech recognition is facilitated by using a regional template in a regional language suitable for him / her. In addition, the voice recognition rate can be increased, and the operability can be improved by reducing the number of inputs in voice recognition.

  The speech recognition method of the present invention is characterized in that each of the regional templates includes a plurality of templates for men, women, and a mixture.

  With this configuration, the user can further increase the recognition rate by selecting a recognition template for men, women, or a mixture according to his / her gender.

  The speech recognition method of the present invention is characterized in that switching of the recognition template is performed according to a user instruction.

  With this configuration, the recognition control unit selects a recognition template according to the selected selection item by selecting in advance from the selection items which recognition template is used for voice recognition. be able to.

  The speech recognition method of the present invention is characterized in that a recognition template having a high recognition rate used last time is stored and used as the next recognition template.

  With this configuration, the user selects which selection item, and the selected regional template stores which regional template is for men, women, or mixed, and the region with the highest recognition rate By storing the template, it is possible to use the same regional template as that used last time when the speech recognition apparatus is activated from the next time, and the recognition efficiency and operability can be improved.

  The present invention also provides a current position detecting means for detecting a current position, a display means for displaying a map based on road map data and a current position detected by the current position detecting means, and claims 1 to 8. An in-vehicle navigation device comprising: the voice recognition device according to claim 1; and a control unit that performs route guidance based on a recognition result of the voice recognition device.

With this configuration, the user can perform an input operation in the navigation device by voice using a regional language suitable for himself, and can easily perform voice input and increase the voice recognition rate. Further, the in-vehicle navigation device of the present invention includes a storage unit that stores a standard phrase and a plurality of regional phrases as a voice guidance phrase, and is used in the voice recognition device. The voice guidance phrase is selected according to the recognized recognition template.

  With this configuration, voice guidance at intersections, highway entrances and the like can be performed in the same local language as the regional template used for voice recognition, and information can be easily transmitted to the user.

  The present invention includes a standard template and a plurality of regional templates as recognition templates used for speech recognition, and includes a recognition control unit that can arbitrarily switch the recognition templates. By selecting and performing voice recognition, voice input can be performed easily, the voice recognition rate can be increased, and convenience and operability can be improved.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 shows the configuration of an in-vehicle navigation device according to an embodiment of the present invention. In FIG. 1, the direction sensor 1 uses a vibration gyro and detects the traveling direction of the own vehicle. The vehicle speed sensor 2 generates a vehicle speed pulse corresponding to the number of rotations of the wheel of the vehicle on which the present apparatus is mounted. Various sensor signals 3 are signals such as a reverse switch, a parking switch, and a light switch. The sensor signal processing unit 4 calculates the traveling direction of the vehicle based on the signal from the direction sensor 1, calculates the travel distance from the vehicle speed signal from the vehicle speed sensor 2, and further requires control based on various sensor signals 3. A simple signal. The DVD-ROM 5 is one of storage means, and includes map data, function word dictionary corresponding to a standard word in voice recognition and a plurality of regional languages, facility name, address, surname, name, and other dictionary data and voice phrase data. Etc. are recorded. The DVD-ROM drive 6 reads map data, voice recognition dictionary data, voice data, and the like 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 includes an operation input unit such as a touch panel on the front surface. The GPS receiver 8 can obtain the current position (latitude / longitude) of the vehicle by receiving and calculating radio waves transmitted from a plurality of satellites through an antenna. The ETC in-vehicle device 9 automatically pays the toll road usage fee wirelessly with the roadside devices installed at the entrance gate and the exit gate of the toll road, and has the credit function inserted in the ETC on-vehicle device 9. A card reader is provided for the payment of charges by using an IC card or a prepaid IC card. In order to ensure security, this card reader is configured to read the content of an inserted IC card type driver's license and confirm the identity. These DVD-ROM drive 6, liquid crystal display 7, GPS receiver 8, ETC vehicle-mounted device 9, etc. are arranged on the dashboard of the vehicle and are connected to the communication interface 12 of the apparatus main body 11 through the in-vehicle LAN 10. The apparatus main body 11 is installed in a trunk room of a vehicle, a center console in a vehicle, or under a seat.

  The microphone 13 is arranged in the vicinity of the driver in the vehicle and inputs a spoken phrase from the user. The speaker 14 is a search result, a voice recognition result, intersection guidance on the travel route, branch guidance, toll gate guidance. It is used for voice guidance such as exit guidance and voice operation instructions on the remote controller. The voice analysis unit 15 performs frequency analysis on the words of the phrase input from the microphone 13 and outputs a time series of a short time spectrum. The image processor 16 performs a display image forming process based on map data, current position data of the own vehicle, building data, and the like. The storage unit 17 is one of storage units, and includes a ROM storing a program, a RAM temporarily storing work data, a VRAM storing image data, and the like, and also functions as a dictionary memory. The recognition template storage unit 18 stores three recognition templates for men, women, and a mixture consisting of short-time spectra of phonemes. The voice processor 19 converts a phoneme symbol sequence output as a voice recognition result into a voice signal, or outputs a search result, voice guidance on the travel route, and a voice signal representing the operation content on the remote controller to the speaker 14. The CPU (central processing unit) 20 controls the entire apparatus, and software programs such as a current position calculation unit, a route search unit, a search unit, and a point registration unit are installed in order to realize a function as a navigation device. At the same time, software programs such as the kana CV conversion unit 21, the dictionary generation unit 22, and the similarity determination unit 23 are executed in order to realize the function as the voice recognition device.

  Next, the operation of this embodiment will be described first. In FIG. 1, when the apparatus is started up by a predetermined operation, the current position detection means in the CPU 20 receives position information from the GPS receiver 8 and signals from the direction sensor 1 and the vehicle speed sensor 2 by the sensor signal processing unit 4. Based on the processed data, the current current position of the vehicle is calculated. Based on the vehicle position information, the CPU 20 reads the corresponding map data from the DVD-ROM 5 through the DVD-ROM drive 6, converts it into image data by the image processor 16, and temporarily stores it in the VRAM of the storage unit 17. The signal is converted into a signal and displayed on the screen of the liquid crystal display 7 together with the vehicle position through the communication interface 12. When an address name such as a destination is input through the microphone 13, the address name is recognized by the voice recognition function, and the destination is set. The route search means in the CPU 20 calculates the optimum guide route from the current position of the vehicle to the specified destination and displays it on the map of the liquid crystal display 7 in a superimposed manner. When the driver advances the vehicle along the guidance route displayed on the liquid crystal display 7, the CPU 20 displays the information on the liquid crystal display 7 on the basis of the current position information calculated by the current position detecting means and the road network data on the map data. The vehicle position mark will be updated sequentially. When the vehicle reaches a branch point in the guidance route, the voice phrase guidance added to the map data is output from the speaker 14. Furthermore, the point to be stored can be registered by the point registration means, and the destination and home address can also be registered. Also, the facility can be searched by genre, 50 sounds, etc. by the search means.

  Next, the speech recognition apparatus in the present embodiment will be described. The speech recognition apparatus according to the present embodiment includes a speech analysis unit 15 that frequency-analyzes the speech of a word input from the microphone 13 and outputs a time series (LPC cepstrum coefficient sequence) of a short-time spectrum, The Kana CV conversion unit 21 that performs CV conversion, the recognition template storage unit 18 that stores three recognition templates for men, women, and a mixture consisting of short-time spectra of phonemes, and the Kana CV conversion word by the recognition template A dictionary generation unit 22 that creates a word dictionary by converting to a phoneme symbol sequence (LPC cepstrum coefficient sequence), a time series of short-term spectra of words output from the speech analysis unit 15, and words generated by the dictionary generation unit 22 The similarity between the phoneme symbol sequence and the phoneme symbol sequence with the smallest distance between them, that is, the highest similarity is used as the recognition result. When the similarity determination unit 23 to output, the speech processor 19 that outputs the phoneme symbol sequence of the recognition result as speech from the speaker 14, and the erroneous recognition that is different from the speech of the word to which the output speech of the recognition result is input The main constituent requirement is the CPU 20 which is also a recognition control means for allowing the user to redo the operation up to three times.

  FIG. 2 shows an example of a recognition template stored in the recognition template storage unit 18 in the present embodiment. The recognition template includes a standard template 31 corresponding to the standard language and a regional template 32 corresponding to the regional language. Each template includes “mixed sex” “male 1” “male 2” “female 1” “female 2”. ”Templates are prepared. “Mixed” is a wide range of voice recognition that can be adapted to anyone who speaks when both men and women are in the vehicle. “Men 1” is for men with high voice. “Men 2” is for men with low voice, “Woman 1” is for women with high voice, and “Woman 2” is for women with low voice. The regional template 32 is roughly categorized for each regional block such as Tohoku dialect, Chubu dialect, Kansai dialect, Kyushu dialect, Okinawa dialect, etc. If you want to increase the recognition rate, create a regional template for each prefecture. Good.

  Next, speech recognition processing using a regional template will be described with reference to FIGS. First, the activation process will be described. In FIG. 3, when “Voice Control” is selected from the menu screen displayed on the liquid crystal display 7, voice recognition preparation is started (step S 1), and voice recognition dictionary data is expanded from the DVD-ROM 5 and sorted in the order of “Aiueo”. Once stored in the dictionary memory of the storage unit 17 (step S2). When the voice recognition preparation is completed (step S3), the user presses the voice input button on the remote controller and speaks into the microphone 13 to capture the voice (step S4), and shifts to the regional template switching process of FIG. To do.

  In the regional template switching process of FIG. 4, first, it is checked whether or not the regional template has been switched by the user (step S5), and what type of switching has been performed (step S6). If the switched method is the area code designation, the selected area is designated (step S7). In this case, for example, by selecting any one of the regional codes assigned to the prefectures in Japan, the regional template for the region can be selected. If the switched method is the current location tracking, the current location area is designated from the road map data read from the DVD-ROM 5 based on the current location information received by the GPS receiver 8 (step S8). If the switched system is designated as a permanent address, specify the area from the driver's license inserted in the card reader of the ETC on-board unit 9 or the permanent address or address read from the IC card (step S9), or point registration A region is designated from the current address registered by the means (step S10). When the region is designated in this way, the CPU 20 switches the region template to the designated region (step S11), and proceeds to the region template selection process of FIG. If it is not switched to the regional template in step S5, it is switched to the standard template (step S12).

  In the regional template selection process of FIG. 5, it is first checked whether the template used in the previous speech recognition process is the default template (step S13). If so, the template set as the default is selected (step S13). S18). Otherwise, it is checked whether the priority template is “mixed sex” (step S14). The priority template is set arbitrarily by the user based on his / her gender and voice quality. If it is “Mixed”, “Mixed” is selected as a template (step S19). If not, it is checked whether the priority template is “male 1” (step S15). If so, “male 1” is selected as the template (step S20). If not, it is checked whether the priority template is “male 2” (step S16). If so, “male 2” is selected as the template (step S21). If not, it is checked whether the priority template is “female 1” (step S17). If so, “female 1” is selected as a template (step S22). Otherwise, “woman 2” is selected as a template (step S23). When the area template is selected in this way, the process proceeds to the determination process of FIG.

  In the determination process of FIG. 6, the voice data captured in step S4 of FIG. 3 is compared with the word data of the regional template selected in the selection process of FIG. 5 (step S24). This corresponds to voice data such as “facility search”, “destination setting”, “point registration”, “current location”, “scroll”, and voice data such as facility name, address, surname, name, etc. The word data to be compared is compared. The word data read from the DVD-ROM 5 is CV converted by the Kana CV conversion unit 21, and then converted into an LPC cepstrum coefficient sequence by the selected region template in the dictionary generation unit 22 to generate a word dictionary, and the storage unit 17 dictionary memories. The words uttered by the user are subjected to frequency analysis by the voice analysis unit 15, converted into an LPC cepstrum coefficient sequence, and output. The similarity determination unit 23 calculates the similarity between the LPC cepstrum coefficient sequence of the input speech and the LPC cepstrum coefficient sequence in the dictionary memory, and selects the one with the highest similarity from the dictionary memory as a recognition result. Output. In this way, the speech recognition dictionary is searched (step S25), and it is checked whether or not the input speech and a word in the dictionary memory have been hit (step S26). The cepstrum coefficient sequence is output to the audio processor 19, and is synthesized with the VCV audio signal by the audio processor 19, and is output as audio from the speaker 14 as talkback (step S27). At the same time, the CPU 20 performs a control operation corresponding to it, and proceeds to the default setting process of FIG. If no hit is found in step S26, retry is allowed up to 3 times (step S28). If successful in the second time, the process returns to the startup process of FIG. 3 and waits for the next voice input, and the third time also fails. In such a case, the voice recognition process is terminated as an error, and a message to that effect is output from the speaker 14. If there is no hit in step S26, the probability of hitting is increased by switching the recognition template to another, for example, switching from “male 1” to “male 2” or “mixed”.

  Next, the learning function in this speech recognition apparatus will be described. As described above, in order to improve the accuracy of speech recognition, it is necessary to select a regional template that matches the user's voice quality. It is possible to improve the recognition rate by storing it and setting the selected regional template as a default in the next speech recognition.

  In the default setting process of FIG. 7, first, a matching comparison process is performed to check whether or not the selected area template has been hit (step S29). First, the “Mixed” template is scored in a five-step evaluation, for example, “Bad”, “Slightly bad”, “Normal”, “Slightly good”, and “Good”, followed by “Men 1”, “ Sequential comparison and scoring are made for male 2 "" woman 1 "" woman 2 "(step S30 to step S34). Then, the template with the highest score is set as the default template from all the scoring results, and the template is adopted from the next speech recognition (step S35).

  In the voice recognition process described above, when a regional template is selected, voice guidance in a regional language that matches the selected regional template may be performed. This is because, in addition to the standard language, a voice guidance phrase in the local language is stored in the DVD-ROM 5 by ADPCM, and when the regional template is selected during the voice recognition processing, it matches the selected regional template. The voice guidance phrase in the local language is read from the DVD-ROM 5, decoded into a voice signal by the voice processor 19, and output from the speaker 14. Thereby, voice guidance at intersections, highway entrances and the like can be performed in the same local language as the local template used for voice recognition, and the user can be relieved.

  As described above, according to the present embodiment, the result of frequency analysis of the voice of the input word is collated with a word dictionary created using a plurality of recognition templates, and a word having a high similarity is extracted and the voice is extracted. When recognizing an image, a standard template and a plurality of regional templates are provided as recognition templates, and the recognition template is arbitrarily switched. Therefore, the user can select a regional template suitable for himself and perform speech recognition. In addition, voice recognition is possible even when a local language is used, and convenience and operability can be improved.

  As described above, the voice recognition device and method and the vehicle-mounted navigation device according to the present invention can recognize a local language suitable for a person even if the accent, intonation, or wording is different, or a pronounced user pronounces it. It has the effect that the recognition rate can be improved by using a template, is useful as a control device that operates by voice input, etc., and is also applicable to applications such as mobile phones, personal computers, robots, and home appliances can do.

Schematic block diagram showing a configuration of an in-vehicle navigation device provided with a voice recognition device in an embodiment of the present invention Schematic diagram showing a recognition template in an embodiment of the present invention The flowchart which shows the speech recognition starting process in embodiment of this invention The flowchart which shows the speech recognition area template switching process in embodiment of this invention. The flowchart which shows the speech recognition area template selection process in embodiment of this invention. The flowchart which shows the speech recognition determination process in embodiment of this invention The flowchart which shows the speech recognition default setting process in embodiment of this invention

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Direction sensor 2 Vehicle speed sensor 3 Various sensor signals 4 Sensor signal processing part 5 DVD-ROM
6 DVD-ROM drive 7 Liquid crystal display 8 GPS receiver 9 ETC in-vehicle device 10 Car LAN
DESCRIPTION OF SYMBOLS 11 Apparatus main body 12 Communication interface 13 Microphone 14 Speaker 15 Voice analysis part 16 Image processor 17 Memory | storage part 18 Recognition template storage part 19 Voice processor 20 CPU
21 Kana CV conversion unit 22 Dictionary generation unit 23 Similarity determination unit

Claims (14)

  A speech recognition apparatus for recognizing speech by extracting a word with high similarity by comparing a result of frequency analysis of speech of an input word with a word dictionary created using a plurality of recognition templates, A speech recognition apparatus comprising a standard template and a plurality of regional templates as templates, and a recognition control means capable of arbitrarily switching the recognition templates.   The speech recognition apparatus according to claim 1, wherein the recognition template includes a plurality of templates for men, women, and a mixture.   The speech recognition apparatus according to claim 1, wherein the recognition control unit switches the recognition template based on a selection item designated by a user.   4. The speech recognition apparatus according to claim 3, wherein the selection item is regional data selected from a regional code displayed on a display means.   4. The speech recognition apparatus according to claim 3, wherein the selection item is a license or IC card data read by a card reading means.   4. The voice recognition apparatus according to claim 3, wherein the selection item is current position data from a current position detecting means for detecting a current position.   4. The speech recognition apparatus according to claim 3, wherein the selection item is home position data registered by a point registration unit.   The said recognition control means is provided with the learning function which memorize | stores the recognition template with the high recognition rate used last time, and uses it as a next recognition template. Voice recognition device.   A method for recognizing speech by extracting a word having a high similarity by comparing a result of frequency analysis of speech of an input word with a word dictionary created using a plurality of recognition templates, and using the standard as the recognition template A speech recognition method comprising a template and a plurality of regional templates, and performing speech recognition by arbitrarily switching the recognition templates.   The speech recognition method according to claim 9, wherein the recognition template includes a plurality of templates for men, women, and a mixture.   The speech recognition method according to claim 9 or 10, wherein switching of the recognition template is performed according to a user instruction.   The speech recognition method according to any one of claims 9 to 11, wherein a recognition template having a high recognition rate used last time is stored and used as a next recognition template.   9. A current position detecting means for detecting a current position, a display means for displaying a map and a current position detected by the current position detecting means on the basis of road map data, and any one of claims 1 to 8. A vehicle-mounted navigation device comprising a voice recognition device and control means for performing route guidance based on a recognition result of the voice recognition device.   The storage device for storing a standard phrase and a plurality of regional phrases as a voice guidance phrase, wherein the voice guidance phrase is selected according to a recognition template used in the voice recognition device. Car navigation system.

Images