JP2018155997A - Apparatus for retrieving facilities - Google Patents

Abstract

To provide a facility search apparatus capable of improving the recognition rate for a facility name. A voice input unit of a navigation device performs voice recognition processing on a facility name uttered by a user and calculates a similarity score corresponding to each of a plurality of facility names as a pronunciation score. The likelihood corresponding to each of a plurality of facility names based on the store number score calculated based on the calculation unit 46 and the number of stores of the chain store as attribute information corresponding to each of the plurality of facility names and the pronunciation score. A total score calculation unit 50 that calculates a total score as a degree, and a recognition result output unit 52 that outputs a voice recognition result corresponding to the facility name uttered by the user based on the calculated total score. Yes. [Selection] Figure 1

Description

  The present invention relates to a facility search apparatus that is mounted on a vehicle or the like and extracts a corresponding facility by voice input.

  Conventionally, after extracting features for place names spoken by the user and calculating the likelihood for multiple recognition target words registered in the speech recognition dictionary, information such as street names corresponding to the lower hierarchy of place names There is known a voice recognition apparatus that obtains a voice recognition result by multiplying a weighting coefficient set based on a quantity to calculate a weighted likelihood (see, for example, Patent Document 1). For example, when a user utters a city name to which many streets belong, since a weighted likelihood higher than the original likelihood corresponding to this city name is calculated, this city name is used as a speech recognition result. The chances of getting higher.

Japanese Patent No. 5199391

  By the way, in the speech recognition apparatus disclosed in Patent Document 1 described above, place names having a tree structure indicating a hierarchical relationship are targeted for speech recognition, and speech recognition using a weighting coefficient for names that do not have a tree structure. There was a problem that processing could not be performed. For example, considering the case where speech recognition processing is performed for a facility name such as a store name uttered by a user, generally, the facility name does not have a tree structure, so speech recognition processing using a weighting coefficient is performed. The recognition rate cannot be improved.

  The present invention has been created in view of such a point, and an object thereof is to provide a facility search apparatus capable of improving the recognition rate of facility names.

  In order to solve the above-described problem, the facility search apparatus of the present invention performs a speech recognition process on a facility name uttered by a user, and calculates a similarity corresponding to each of a plurality of recognition target vocabularies. A likelihood calculating means, a likelihood calculating means for calculating a likelihood corresponding to each of the plurality of recognition target words based on the attribute information and similarity corresponding to each of the plurality of recognition target words, and a likelihood calculating means And a recognition result output means for outputting a speech recognition result corresponding to the facility name uttered by the user based on the likelihood calculated by the above.

  In addition to the similarity between the name of the facility that spoke and multiple vocabulary to be recognized, attribute information corresponding to each vocabulary to be recognized is combined between multiple vocabulary to be recognized separately from the speech recognition results. Therefore, the facility name recognition rate can be improved as compared with the case where the recognition result is obtained based only on the voice uttered by the user.

  In addition, the recognition result output unit described above is arranged such that the facility name indicated by the predetermined number of recognition target words in the descending order of the likelihood calculated by the likelihood calculating unit is the highest likelihood facility name. In addition, it is desirable to display it as a voice recognition result. Thereby, since the possibility that the name of the facility arranged at the top is correct as the recognition result is high, the correct facility name is most easily noticeable, and the operation for selecting this facility name is facilitated.

  Further, the above-described recognition result output means divides the facility name indicated by a predetermined number of recognition target words into a plurality of pages, and displays the speech recognition result so that the most likely facility name is included in the front page. Is desirable. As a result, the facility name with the highest likelihood is not placed on the second and subsequent pages, and a complicated operation such as page turning for selecting the facility name becomes unnecessary.

  The likelihood calculating means described above calculates the likelihood by adding a predetermined weight to each of the first value corresponding to the similarity and the second value corresponding to the attribute information. Is desirable. Thereby, the contribution degree by the attribute information in likelihood can be adjusted, and a specific facility can be obtained preferentially as a voice recognition result.

  Moreover, it is desirable that the attribute information described above is information correlated to the degree of familiarity of the facility name corresponding to each of the plurality of recognition target vocabularies. Alternatively, the attribute information described above is preferably information indicating the number of stores belonging to a chain store when the facility corresponding to each of the plurality of recognition target vocabularies is a store of the chain store. Alternatively, the attribute information described above is information that indicates that facilities corresponding to each of a plurality of recognition target vocabularies are famous nationwide, and / or is famous in any of a plurality of regions that divide the country into a plurality of regions. It is desirable that the information indicates that it is. In general, when there are multiple facilities similar to the voice spoken by the user, it is based on the name recognition (in the case of a chain store with a large number of stores or when it is known to be famous in the whole country or one region). In addition, since the likelihood can be increased, the recognition rate can be improved.

It is a figure which shows the detailed structure of the navigation apparatus of one Embodiment. It is a flowchart which shows the operation | movement procedure which performs the speech recognition process with respect to an input audio | voice and extracts one facility. It is a figure which shows the specific example of the comprehensive score calculated based on the pronunciation score and the store number score. It is a figure which shows the example of a display of a recognition result. It is a figure which shows the specific example of the total score computed based on the pronunciation score and the famousness score. It is a figure which shows the specific example of the total score computed based on the pronunciation score, the store number score, and the famousness score.

  Hereinafter, a navigation device according to an embodiment to which a facility search device of the present invention is applied will be described with reference to the drawings.

  FIG. 1 is a diagram illustrating a detailed configuration of a navigation device 100 according to an embodiment. As shown in FIG. 1, the navigation device 100 includes a navigation controller 1, a map data storage device 2, an operation unit 3, a vehicle position detection unit 4, a display device 5, an audio unit 6, and a microphone 7. This navigation device 100 is mounted on a vehicle. .

  The navigation controller 1 executes a predetermined operation program using a CPU, ROM, RAM, etc., thereby setting a map image display operation around the vehicle position, a facility search operation, and a travel route connecting the departure point and the destination. Various functions such as a route search process to be performed and a route guidance operation for guiding the traveling of the vehicle along the traveling route are realized. The detailed configuration of the navigation controller 1 will be described later.

  The map data storage device 2 is a storage medium storing map data and a reading device thereof. This map data includes map data used for map display, route search / guidance, search data necessary for facility search, and the like. This search data includes a speech recognition dictionary necessary for speech recognition processing when inputting a facility name by voice, a facility DB (database) having facility names, locations, categories, and other detailed information for each facility. include.

  The map data storage device 2 stores map data in units of rectangular figure leaves divided into appropriate sizes by longitude and latitude. The map data of each leaf can be specified and read by designating the leaf number. The map data storage device 2 is realized by a hard disk device or a semiconductor memory, or by a DVD and its reading device. Further, the map data storage device 2 may be replaced with a communication device to acquire map data from an external map distribution server (not shown).

  The operation unit 3 is for receiving user instructions (operations), and includes various operation buttons and operation knobs. In addition, the operation unit 3 includes a touch panel attached to the screen of the display device 5, and an operation instruction can be performed when a user directly points a part on the screen with a finger or the like. Yes. The vehicle position detection unit 4 includes, for example, a GPS receiver, an orientation sensor, a distance sensor, and the like, detects the vehicle position (longitude, latitude) at a predetermined timing, and outputs the detection result.

  The display device 5 is configured by, for example, an LCD (liquid crystal display device), and displays a map image around the vehicle position based on the video signal output from the navigation controller 1. The audio unit 6 outputs a guidance voice or the like generated based on the voice signal input from the navigation controller 1 into the vehicle interior. The microphone 7 is used to input a voice such as a facility name, and collects a user's voice.

  Next, a detailed configuration of the navigation controller 1 will be described. The navigation controller 1 shown in FIG. 1 includes a map buffer 10, a map readout control unit 12, a map drawing unit 14, a vehicle position calculation unit 20, a route search processing unit 30, a route guidance processing unit 32, a destination setting unit 34, and a facility search. Unit 40, input processing unit 60, and display processing unit 70.

  The map buffer 10 temporarily stores map data and search data read from the map data storage device 2. This map data includes at least data necessary for drawing a map image and data necessary for route search and route guidance. The map reading control unit 12 sends a map data reading request within a predetermined range in accordance with the vehicle position calculated by the vehicle position calculating unit 20 or the position designated by the user by operating the operation unit 3. Output to. Based on the normal map data stored in the map buffer 10, the map drawing unit 14 performs drawing processing necessary to display a map image on the display device 5 and creates map drawing data.

  The vehicle position calculation unit 20 calculates the vehicle position based on the detection data output from the vehicle position detection unit 4, and if the calculated vehicle position is not on the road of the map data, the vehicle position calculation unit 20 calculates the vehicle position. Perform map matching processing to correct.

  The route search processing unit 30 calculates a travel route (guidance route) having a minimum cost from the departure point to the destination by route search processing using a predetermined search condition. The route guidance processing unit 32 creates guidance route drawing data for displaying the travel route obtained by the search processing by the route search processing unit 30 so as to be superimposed on the map or displaying the guidance map of the right-left turn intersection. A voice signal such as intersection guidance necessary for guiding the vehicle to travel along the travel route is generated (route guidance operation). The destination setting unit 34 sets a destination used for the route search processing performed by the route search processing unit 30.

  The facility search unit 40 extracts a facility designated by the user and searches for detailed information on the extracted facility. Here, it is assumed that the facility is designated by the user by voice input. The facility search unit 40 includes a voice input unit 42, a facility selection unit 54, a facility DB (database) 56, and a detailed information search unit 58.

  When the user inputs a facility name by voice, the voice input unit 42 performs voice recognition processing on the input voice and extracts a plurality of candidate facility names. The extracted plurality of facility names are input to the display processing unit 70 as a voice recognition result and displayed on the display device 5. The facility selection unit 54 selects one from a plurality of facility names extracted by the voice input unit 42 according to a user instruction. The facility DB 56 stores detailed information on facilities. For example, the category (industry) of the facility, the business contents, etc., the chain ID for identifying the chain store if it is a store of a chain store, the facility is famous nationwide or famous in some regions (For example, “1” is set when famous nationwide, “2” is set when famous in some regions, and “0” is set otherwise) ) Is included in the detailed information of the facility. The detailed information search unit 58 searches for detailed information on the facility corresponding to the facility name selected by the facility selecting unit 54 using the map DB 56.

  Further, the voice input unit 42 described above includes a voice recognition dictionary 44, a pronunciation score calculation unit 46, a store number score calculation unit 48, a total score calculation unit 50, and a recognition result output unit 52.

  The speech recognition dictionary 44 includes a facility name and how to read the facility name (phoneme model) for each of a plurality of facilities (recognition target vocabulary) registered in advance. The pronunciation score calculation unit 46 extracts a feature amount from the facility name by the user's utterance collected by the microphone 7, and the feature amount and the feature amount of the phoneme model of each facility stored in the speech recognition dictionary 44. And the similarity for each facility is calculated as a pronunciation score. For example, assuming that the pronunciation score when the feature quantity of the facility name inputted by speech and the feature quantity of the phoneme model corresponding to any of the facilities stored in the speech recognition dictionary 44 completely match is 100, A pronunciation score of less than 100 is obtained according to the similarity.

  The store number score calculation unit 48 calculates a store number score corresponding to the scale (number of stores) of the chain store as attribute information of each facility for each of the plurality of facilities whose pronunciation score is calculated by the pronunciation score calculation unit 46. To do. For example, the scale of a chain store is divided into 10 stages, 100 if the chain belongs to the largest chain store, 90 if it belongs to the chain store of the next scale, and so on. 70,..., 10, and 0 may not be a chain store. Note that such a method for setting the store number score is an example, and the method of dividing the store number score is set to other than 10 stages, the maximum value of the store number score is set to other than 100 (for example, 50), or appropriately changed. May be. Further, the number of stores of chain stores to which each facility belongs can be known by counting the same one from the chain IDs of each facility stored in the facility DB 56, but the stores of chain stores belonging to each facility The numbers may be stored in the facility DB 56 in correspondence with each other.

  Based on the pronunciation score calculated by the pronunciation score calculation unit 46 and the store number score calculated by the store number score calculation unit 48, the general score calculation unit 50 calculates the pronunciation score by the pronunciation score calculation unit 46. A total score as a likelihood is calculated for each of a plurality of facilities. For example, when the pronunciation score is the first value A and the store number score is the second value B, the total score C is obtained by adding a predetermined weight to each value (C = aA + bB). It is possible to calculate. If each of a and b is 0.5 (50%), C = 0.5A + 0.5B, and the total score C is calculated by a simple arithmetic mean. Further, the values of a and b may be other than 0.5. Alternatively, instead of performing such weighting, the total cost C may be calculated by a geometric mean (C = √ (A × B)) or other operations.

  The recognition result output unit 52 extracts a plurality of (for example, 10) facility names in descending order of the total score calculated by the total score calculation unit 50, and outputs the result as a recognition result. For example, the recognition result output unit 52 divides a plurality of facility names into a group with a high overall score and a group with a low overall score, and creates recognition result image drawing data indicating the recognition results for two pages.

  The input processing unit 60 outputs commands for performing operations corresponding to various operation instructions input from the operation unit 3 toward each unit in the navigation controller 1. The display processing unit 70 receives normal map drawing data created by the map drawing unit 14 and displays a normal map in a predetermined range on the screen of the display device 5 based on the drawing data. In addition, when guidance route drawing data indicating a normal driving route or the like created by the route search processing unit 30 is input, the display processing unit 70 guides a driving route corresponding to the drawing data or a right / left turn intersection. Is superimposed on the map and displayed on the screen of the display device 5. When the recognition result image drawing data created by the recognition result output unit 52 is input, the display processing unit 70 displays a recognition result including a plurality of facility names. This recognition result is created for two pages, and a first recognition result list including a group with a high overall score is displayed at first. However, when the page is directed by the user, a group with a low overall score is displayed. The display is switched to the second recognition result list including.

  The pronunciation score calculation unit 46 described above corresponds to the similarity calculation unit, the store number score calculation unit 48, the total score calculation unit 50 corresponds to the likelihood calculation unit, and the recognition result output unit 52 corresponds to the recognition result output unit.

  The navigation apparatus 100 of this embodiment has such a configuration, and next, the operation thereof will be described. FIG. 2 is a flowchart showing an operation procedure for performing voice recognition processing on an input voice and extracting one facility.

  When facility search is instructed by the user, speech recognition processing for the facility name by the speech input unit 42 is started. When the facility name is input by voice (step 100), the pronunciation score calculation unit 46 extracts the feature amount for the facility name input by the speech, and the pronunciation score for a plurality of facilities in the order in which the feature amounts of the phoneme model are similar. Is calculated (step 102). In addition, the store number score calculation unit 48 calculates the store number score for a plurality of facilities whose pronunciation score is to be calculated by the pronunciation score calculation unit 46 (step 104).

  Next, the total score calculation unit 50 calculates a total score using the pronunciation score calculated by the pronunciation score calculation unit 46 and the store number score calculated by the store number score calculation unit 48 (step 106). Note that the above-described calculation of the pronunciation score, the store number score, and the overall score is not performed separately but in parallel.

  FIG. 3 is a diagram illustrating a specific example of the total score calculated based on the pronunciation score and the store number score. In the example shown in FIG. 3, for the facility P1, when the pronunciation score A is 60 and the store number score B is 100, and the weights a and b are both 50%, the total score D is 80. Similarly, for the facility P2, the pronunciation score A is 70, the store number score B is 40, and the total score D is 55. For facility P3, pronunciation score A is 80, store number score B is 30, and total score D is 55.

  Next, the recognition result output unit 52 extracts a plurality of (for example, 10) facilities from the top in the order of higher total score (in descending order) and outputs (displays) the results as recognition results (step 108). .

  FIG. 4 is a diagram illustrating a display example of the recognition result. As shown in FIG. 4, as a recognition result, ten facilities are extracted in descending order of the overall score, and the facility names p1, p2, and p5 of the top five facilities P1, P2, P3, P4, and P5 having the highest total score are shown. The recognition result image D1 of the first page including p3, p4, and p5, and the facility names p6, p7, p8, p9, and p10 of the lower five facilities P6, P7, P8, P9, and P10 having a low overall score are The recognition result image D2 of the second page included is created by the recognition result output unit 52, and the recognition result image D1 of the first page is first displayed on the display device 5. In this state, when the page is instructed by the user (in addition to instructing by using the operation unit 3 or instructing by voice input), the recognition result image D2 of the second page is displayed on the display device. 5 is displayed. In each of the recognition result images D1 and D2, the facilities having a high total score are arranged higher and the facilities are arranged lower as the total score becomes lower.

  In this way, in a state where the recognition result images D1 and D2 are displayed, the facility selection unit 54 determines whether any facility (facility name) has been selected by the user (step 110). If the selection operation is not performed, a negative determination is made and this determination is repeated. When the user performs a selection operation for designating any facility, an affirmative determination is made in the determination in step 110. Next, the facility selection unit 54 selects one facility according to the user's instruction (step 112). In this way, a series of operations related to facility extraction using voice input is completed. The extracted facility can be used as a target facility for detailed information search by the detailed information search unit 58 or can be used as a destination set by the destination setting unit 34.

  As described above, in the voice input unit 42 of the navigation device 100 according to the present embodiment, not only the similarity between the name of the facility spoken and a plurality of recognition target words, but also attribute information (each facility belongs) corresponding to each recognition target word. By combining the number of chain stores), it is possible to make a difference in the degree of selection between a plurality of recognition target vocabularies separately from the speech recognition result, and therefore based only on the speech uttered by the user Thus, the recognition rate can be improved as compared with the case where the recognition result is obtained.

  In addition, a predetermined number of facility names are displayed in order from the highest total score (likelihood) calculated by the total score calculation unit 50 as the voice recognition result so that the facility name with the highest total score is placed at the top. ing. Thereby, since the possibility that the name of the facility arranged at the top is correct as the recognition result is high, the correct facility name is most easily noticeable, and the operation for selecting this facility name is facilitated.

  In addition, since the recognition result is divided into a plurality of pages and the facility name having the highest overall score is displayed in the first page, the facility name having the highest overall score may be arranged on the second and subsequent pages. This eliminates the need for complicated operations such as page turning for selection after the facility name with the highest overall score is displayed.

  In addition, when calculating the total score, each of the pronunciation score and the store number score is weighted and added together, so the contribution by the number of stores of the chain store in the total score is adjusted and the voice A specific facility can be preferentially obtained as a recognition result.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation implementation is possible within the range of the summary of this invention. For example, in the above-described embodiment, the store number score is calculated using the number of stores of the chain store that has a correlation with the name recognition of each facility as the attribute information of each facility. However, there is a correlation with the name recognition of each facility. Other attribute information may be used if it is attribute information. For example, a well-known score may be calculated using a well-known flag corresponding to each facility and used instead of the store number score or together with the store number score. Alternatively, the use history of each facility may be recorded, a use score having a large value may be set for a facility having a large number of uses, and used instead of or together with the store number score or the name recognition score.

  FIG. 5 is a diagram illustrating a specific example of the total score calculated based on the pronunciation score and the familiarity score. In the example shown in FIG. 5, for the facility P1, the pronunciation score A is 60 and the name recognition score C is 100 (when the name recognition flag is “1” (nationally famous)), and each of the weights a and c is 50. In the case of%, the total score D is 80. Similarly, for the facility P2, the pronunciation score A is 70, the name recognition score C is 50 (when the name recognition score is “2” (famous in some regions), and the total score D is 60. For the facility P3, the pronunciation score A is 80, the name recognition score C is 0 (when the name recognition score is “0”), and the total score D is 40.

  FIG. 6 is a diagram illustrating a specific example of the total score calculated based on the pronunciation score, the store number score, and the name recognition score. In the example shown in FIG. 6, for the facility P1, the pronunciation score A is 60, the number-of-stores score B is 100, and the visibility score C is 100, and the weights a, b, and c are 50%, 30%, and 20%, respectively. In this case, the total score D is 80. Similarly, for the facility P2, the pronunciation score A is 70, the store number score B is 40, the name recognition score C is 50, and the total score D is 57. For the facility P3, the pronunciation score A is 80, the number-of-stores score B is 30, the visibility score C is 0, and the total score D is 49.

  In this way, when there are a plurality of facilities similar to the voice uttered by the user, the total score is increased based on the name recognition by using the name recognition score instead of the store number score or together with the number of stores score. Therefore, the recognition rate can be improved by preferentially obtaining the name of a facility with a known degree as a recognition result.

  In the above-described embodiment, the case where the facility search unit 40 and the voice input unit 42 are provided in a part of the navigation device 100 has been described. However, the facility search unit 40 and the voice input unit 42 are separate from the navigation device 100. The facility name as a recognition result and the detailed information of the facility as a search result may be input to the navigation device 100. Further, the facility search unit 40 and the voice input unit 42 may be combined with devices other than the navigation device 100 or may be used alone.

  As described above, according to the present invention, not only the degree of similarity between the name of a facility spoken and a plurality of recognition target vocabularies, but also a combination of attribute information corresponding to each recognition target vocabulary, In addition, since the degree of selection can be made different from the recognition result by voice, the recognition rate can be improved as compared with the case of obtaining the recognition result based only on the voice uttered by the user.

DESCRIPTION OF SYMBOLS 1 Navigation controller 10 Map buffer 40 Facility search part 42 Voice input part 44 Speech recognition dictionary 46 Pronunciation score calculation part 48 Store number score calculation part 50 Comprehensive score calculation part 52 Recognition result output part 54 Facility selection part 56 Facility DB (database)
58 Detailed Information Search Unit 60 Input Processing Unit 70 Display Processing Unit 100 Navigation Device

Claims (7)

A similarity calculation means for performing voice recognition processing on the facility name uttered by the user and calculating a similarity corresponding to each of a plurality of recognition target vocabularies;
A likelihood calculating means for calculating a likelihood corresponding to each of the plurality of recognition target words based on the attribute information corresponding to each of the plurality of recognition target words and the similarity;
A recognition result output means for outputting a speech recognition result corresponding to the facility name uttered by the user based on the likelihood calculated by the likelihood calculating means;
A facility retrieval apparatus comprising:   The recognition result output unit arranges the facility names indicated by a predetermined number of the recognition target words in descending order of the likelihood calculated by the likelihood calculating unit so that the facility name having the highest likelihood is arranged at the top. The facility search device according to claim 1, wherein the facility search device is displayed as a voice recognition result.   The recognition result output means divides a facility name indicated by a predetermined number of the vocabulary to be recognized into a plurality of pages, and displays a speech recognition result so that the most likely facility name is included in the front page. The facility search device according to claim 1 or 2.   The likelihood calculating means calculates the likelihood by adding a predetermined weight to each of the first value corresponding to the similarity and the second value corresponding to the attribute information. The facility search device according to any one of claims 1 to 3, wherein:   The facility search apparatus according to any one of claims 1 to 4, wherein the attribute information is information correlated with a degree of familiarity of a facility name corresponding to each of the plurality of recognition target vocabularies.   The attribute information is information indicating the number of stores belonging to a chain store when a facility corresponding to each of the plurality of recognition target vocabularies is a store of the chain store. The facility search device according to any one of the above.   The attribute information is famous in any of the information indicating that the facility corresponding to each of the plurality of recognition target vocabularies is famous nationwide and / or in a plurality of regions divided into a plurality of nations. It is the information which shows that, The facility search apparatus as described in any one of Claims 1-6 characterized by the above-mentioned.

Images