JP2009169470A - Information guidance system and its recognition dictionary database update method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently store a related vocabulary so that desired retrieval can be performed without knowing a name recorded in a recognition dictionary database in retrieval by speech recognition. <P>SOLUTION: In an information guidance system, a speech retrieval part 17 recognizes a keyword, such as a facility name, stored in the recognition dictionary database 13 from a speech signal, and a navigation function part 19 generates guidance information on a route to the facilities based on a recognition result to present it on a speaker display 16. A related vocabulary gathering part 21 of an information center 2 gathers a related keyword having meaning relevance to the initial keyword previously stored in the recognition dictionary database from WEB information. When a navigation device is started, a dictionary update part 18 adds the related keyword gathered by the related vocabulary gathering part to the recognition dictionary database and then the speech retrieval part determine the initial keyword and related keyword as objects to recognized, so that a user can retrieve the desired facilities without knowing the formal name thereof. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an information guidance system for recognizing a keyword stored in a recognition dictionary database by voice recognition and providing guidance information, and a method for updating a recognition dictionary database used therefor.

2. Description of the Related Art Conventionally, a voice interface that recognizes a user's utterance by voice recognition technology and performs information provision and device operation according to the recognition result is known. Such a voice interface is, for example, a car navigation system or a telephone. It is applied to an IVR (Interactive Voice Response) which is a response system.
A service often used in such a system is a facility search.
This system provides services such as navigation to the desired facility location and telephone connection by the user speaking the facility name, the prefecture where the facility is located, the type of facility, etc., and the system recognizes and searches for it. To do.
In such a facility search, there is a problem that there is a lot of information related to each facility and the update frequency is fast, so that the freshness of the information cannot be maintained.

Therefore, for example, Japanese Patent Application Laid-Open No. 2001-289661 proposes a facility that searches for facilities with the latest information by voice search with Internet information so that the facilities can always be searched with the latest information. .
In this case, when the condition of a facility to be searched for is specified by a user, a matching facility is extracted and guided in consideration of the event, fee, weather information, etc. of the facility from the Internet content such as a home page.
Thereby, when the desired purpose is roughly determined, but the facility name is not fixed, an appropriate facility is recommended, and convenience is enhanced.
JP 2001-289661 A

On the other hand, the facility to be searched is already determined among the users, and when searching for the facility that the user desires, it is assumed that the recognition vocabulary described in the recognition dictionary database, that is, the facility name is uttered.
However, this recognition dictionary database is created using names that each facility owner has applied for to a public institution or the like. Therefore, in order for the user to obtain a desired search result, the user needs to correctly speak the facility name recorded on the recognition dictionary database.

For this reason, even if the user remembers the abbreviation of the facility and the characteristic keyword of the facility, if the user does not remember the correct vocabulary (correct facility name), the user can search for the desired facility. Can not.
For example, the temple shrine “Meigetsu-in” in Kamakura is widely known as the “famous temple of hydrangea” and is also known as the “hydrangea temple”. If you know that there is, but you do not know that the temple / shrine's official name is “Meigetsu-in”, you cannot search for the facility.

  Therefore, in view of the above problems, the present invention efficiently updates the recognition dictionary database to accumulate related vocabulary and performs a search desired by the user even when the names recorded in the recognition dictionary database are not known. An object of the present invention is to provide an information guidance system that can be used and a method for updating the recognition dictionary database.

  The present invention collects related keywords that have semantic relevance to the initial keywords stored in advance in the recognition dictionary database from the WEB information, adds the collected related keywords to the recognition dictionary database at a predetermined timing, and performs speech recognition. The number of subjects was increased.

  According to the present invention, even if the user does not know the exact vocabulary, the recognition dictionary database is automatically updated with the related keywords related to the initial keywords. As a recognition target, it is possible to search for information desired by a user.

Next, embodiments of the present invention will be described by way of examples.
FIG. 1 shows an overall configuration of an embodiment in which the present invention is applied to a navigation system.
A client-side navigation device 1 mounted on a vehicle (not shown), an information center 2 accessible from the navigation device 1, and the information center 2 communicate to acquire WEB information (document information on the Internet). A navigation system 5 is constituted by the Internet 3.
In the present embodiment, a case where facility information is provided as navigation information by the navigation device 1 will be described as an example.

First, the navigation device 1 includes a microphone 10, an AD conversion unit 11, a calculation unit 12, a recognition dictionary database (recognition dictionary DB) 13, a communication unit 14, a DA conversion unit 15, a speaker display 16, The current position acquisition unit 50 is configured.
The microphone 10 is for acquiring the voice uttered by the user for searching.
The current position acquisition unit 50 acquires the current position of the host vehicle from a GPS signal received by a GPS (Global Positioning System) signal receiving antenna (not shown).

The AD conversion unit 11 converts the signal acquired by the microphone 10 from an analog signal to a digital signal and outputs the digital signal to the calculation unit 12.
In addition to the voice signal from the user, the microphone 10 may pick up the noise signal generated inside and outside the vehicle, the audio, the sound signal output from the navigation system, etc. It is preferable to provide a means for obtaining with high accuracy.

  For example, (1) a filter is provided between the microphone 10 and the AD conversion unit 11 or between the AD conversion unit 11 and the calculation unit 12 to weaken non-target components in the sound signal. (2) Gain adjustment A mechanism is provided to adjust the amount of amplification so that the power (gain) of the input signal is appropriate. (3) Voice extraction (VAD) that accurately cuts out a section including the voice in the input signal based on the power change of the input signal. : (Voice Activity Detection) mechanism, (4) Echo canceling mechanism for canceling sound signal output from audio system or navigation system, (5) Microphone array using a highly directional microphone or a plurality of microphones For signal enhancement and attenuation processing according to the type and direction of the sound signal. There are also. As a result, noise and input sound that are not subject to speech recognition can be eliminated, and a target speech signal can be acquired preferentially.

The calculation unit 12 includes a voice search unit 17, a dictionary update unit 18, and a navigation function unit 19 by combining general operation circuits such as a CPU, MPU, DSP, and FPGA.
The voice search unit 17 compares and collates a part or all of the uttered voice acquired from the user via the microphone 10 with the vocabulary (keyword) stored in the recognition dictionary database 13, and has a high likelihood of matching. Acquire one or more of the most likely vocabulary as recognition results.

Note that the feature amount can be used for the comparison between the uttered voice and the information stored in the recognition dictionary database 13 in the voice recognition processing by the voice search unit 17. As the feature quantity, time series vector data combining an LPC cepstrum, an LPC delta cepstrum, a mel cepstrum, logarithmic power, and the like is known.
The voice search unit 17 also has a function of monitoring the facility search use history by user's voice and manual operation and recording the use frequency for each region.

The dictionary update unit 18 determines the update timing of related keywords stored in the recognition dictionary database 13 (details will be described later), controls update contents / update ranges, obtains update information using the communication unit 14, and recognizes the dictionary. The database 13 is updated.
The communication network handled by the communication unit 14 includes a mobile phone network, UWB, DSRC, wireless LAN, terrestrial data broadcast, satellite broadcast wave, and the like, and an appropriate one may be selected for communication.

  The navigation function unit 19 has a function of accumulating and analyzing a travel history and the like as a basic function. Then, the voice recognized by the voice search unit 17 is accepted as a destination, the history of the destination is further stored, for example, a registered position input as a waypoint is accepted, and the facility name stored in the recognition dictionary database 13 Are received as destinations and registered positions, and guidance information for guiding the route of the host vehicle is generated based on the current position of the host vehicle acquired by the current position acquisition unit 50.

The recognition dictionary database 13 stores an initial keyword to be recognized and a related keyword that is a vocabulary highly relevant to the initial keyword, together with the degree of relevance, in the speech recognition processing by the speech search unit 17.
As the recognition dictionary database 13, a general storage medium such as a cache memory, main memory, HDD, CD, MD, DVD, optical disk, FDD, or the like is used.

In the recognition dictionary database 13, as shown in FIG. 2, the facility type, longitude / latitude position information, postal code, address, telephone number are recorded in a form accompanying the facility name as an initial keyword. Related keywords are recorded.
In addition, the column of the related keyword of FIG. 2 has shown the state which is not added, and all are NULL.
In the present embodiment, since the stored contents of the recognition dictionary database 13 are changed every time the related keyword is updated, a dictionary version ID indicating the version of the database is recorded for each update.

In the speech recognition processing, speech comparison / collation processing is performed using a recognition dictionary composed of an acoustic dictionary and a language dictionary.
The acoustic dictionary is also called an acoustic model, and records the acoustic feature quantity of a language in a format such as a hidden Markov model (HMM).
One language dictionary is also called a language model, and records the keywords to be recognized as word lists or word connection definitions.

Furthermore, as a language dictionary, a networked language model that explicitly describes the keywords to be recognized and their connection relationships as a network, and analysis of a large amount of text data (text corpus) that shows the probability of occurrence of individual words and the probability of connection between words A statistical probability type language model (statistical model, n-gram) that is automatically determined from the above is known.
In this embodiment, the language model format is not limited, but the network type language model is relatively suitable because the keywords to be recognized are limited to facilities that can be guided. The recognition vocabulary used in the network type language model is expressed in a hierarchical structure from the area name including the destination or the genre of the destination facility to the final destination.

An example of the language model is shown in FIG.
In this example, the network-type language model is configured so that utterances that freely combine items of address prefectures, municipalities, initial keywords, and related keywords in the recognition dictionary database of FIG. 2 can be recognized. ing. In combination with the garbage model (garbage), interjections and rephrasing during speech are skipped.
With this language model, for example, it is possible to recognize complex search conditions such as “XX in Kanagawa”, and by extracting candidates from the recognition dictionary database using the recognition result as a “search key”. It is possible to obtain a desired search result, that is, a facility name.
Similarly, it is possible to generate a language model for recognizing an utterance combining a telephone number and a postal code.

  In this embodiment, the related keyword is updated. Specifically, this corresponds to updating the related keyword shown in FIG. 2 (all are NULL in FIG. 2). Further, when the recognition dictionary database is updated, the related keywords of the language model shown in FIG. 3 (all null values (NULL) in FIG. 3) are also automatically updated.

The speaker display 16 presents the search result by the voice search unit 17 to the user by voice or video.
In addition, after the search result is presented, the navigation function unit 19 further prompts input from the user in accordance with the presented content, and can shift to control according to the input.
For example, when a plurality of facilities are obtained as candidates as a result of the search by the voice search unit 17, the candidates are presented through the speaker display 16 and prompted to be selected by the user. The navigation function unit 19 controls presentation of detailed explanation information of the selected facility, display of a map, route guidance, telephone function, and the like.

The information center 2 includes a communication unit 20, a related vocabulary collection unit 21, and a related keyword database 22.
The communication unit 20 is for communicating with the communication unit 14 of the navigation device 1.
The related vocabulary collection unit 21 can be connected to the Internet 3 and can acquire WEB information 30 existing on the Internet 3.
Specifically, the related vocabulary collection unit 21 performs text analysis using the initial keyword as a key from the WEB information 30, and extracts a related keyword highly relevant to the initial keyword.

The related vocabulary collection unit 21 receives the update request from the dictionary update unit 18 transmitted from the navigation device 1 through the communication unit 20, and updates the related keywords extracted in response to the update request through the communication unit 20 and the communication unit 14. Reply to part 18.
The related vocabulary collection unit 21 is configured to accumulate a search keyword and a record of the reached WEB page when the Internet search is performed manually in the information center.
A detailed related vocabulary collection method of the related vocabulary collection unit 21 will be described later.

The related keyword database 22 stores initial keywords and related keywords associated therewith. This initial keyword is the same as the initial keyword held in the recognition dictionary database 13 described above.
A specific example of data recorded in the related keyword database 22 is shown in FIG. In the example of FIG. 4, in addition to the initial keyword and the related keyword, the facility type and address, the related degree of the related keyword, the version ID, and the like are recorded.
These data can be used to efficiently perform new collection and update processing of related keywords.

Next, as a specific related keyword collection process in the related vocabulary collection unit 21, an example of collecting using text information included in the WEB page itself will be described.
Here, the WEB information 30 on the Internet 3 is searched using the initial keyword registered in the related keyword database 22 shown in FIG. 4 as a search key.
Among the WEB pages of the WEB information 30 acquired as a search result, a predetermined upper page (for example, the upper 20 pages) is set as an analysis target, and the text information in the WEB page is analyzed.
At the time of collecting the first related vocabulary, the data of the related keyword, the degree of association, and the version ID (time stamp) in FIG. 4 are NULL.

Here, what is desired to be analyzed from the WEB page of the WEB information 30 is a related keyword for the facility name which is an initial keyword.
Therefore, it is desirable that the WEB page to be analyzed is a page referring to the facility.
However, there are cases where a WEB page that should not be a target is included in the WEB information of the search result.
For example, if a WEB search is performed using “A temple” as a key in order to extract related keywords for a temple company called “A temple”, initial keywords such as “A temple station” and “A temple high school” The WEB page of another facility to include may correspond. In addition, there may be a case where there is an area name “Y temple A city” in the same name as “A temple”.

If such a WEB page is set as a related keyword analysis target for “A temple”, a related keyword that is not suitable for “A temple” may be extracted.
Therefore, in order to avoid this, the related vocabulary collection unit 21 collects information such as the facility type and address along with the information, and filters (appropriately) the WEB page obtained by the search using the information such as the facility type and address. A process of excluding non-WEB pages from the analysis target).

Specifically, in the case of “A temple”, a WEB page where texts such as “school” and “station”, which are facility types other than “temple”, appear more frequently than a predetermined value is about “A temple”. Excluded from analysis because it is likely not a page to mention.
Similarly, by checking with address information, it is highly likely that a page in which text such as “Y prefecture A city A temple” appears more frequently than the predetermined value is not a page referring to “A temple”. And eliminate.

When the WEB page to be analyzed is determined by the filtering process, the related vocabulary collection unit 21 further analyzes the text information included in the WEB page and extracts a related keyword. As analysis methods, methods used in general morphological analysis and syntax analysis can be applied.
There are several methods for extracting related keywords.
The first method is based on the appearance frequency of words obtained by text analysis.
In other words, the word segmentation process divides sentences into words, sorts parts of speech, erases unnecessary words such as interjections, particles, and conjunctions, and becomes a keyword that leads to facility aliases such as nouns, verbs, and adjectives, and images of facilities. Extract only possible part of speech and calculate the frequency for each word (vocabulary).
Then, as a result of accumulating the word frequencies for all WEB pages to be analyzed, the appearance frequency as a relative value based on the amount of words or the amount of text on all pages is greater than or equal to a predetermined value. Extract words as related keywords.
Here, of course, the initial keyword used as a search key is excluded.

If the appearance frequency is simply counted, it is possible that words that are not suitable as related keywords sometimes appear frequently. Therefore, the extraction accuracy can be increased by combining the following various processes.
For example, business hours of facilities are often listed on the WEB page, and keywords such as “business hours”, “observation hours”, and “XX hours to XX hours” are frequently recorded, and related keywords It may become.
Therefore, it is preferable to prepare such exception vocabulary in advance, and to exclude the exception vocabulary by regularly inspecting and maintaining the exception vocabulary.

As a second method, in order to improve the extraction accuracy, only “words appearing with a specific wording” are counted from the beginning.
Specific phrases include: “speaking of”, “famous (as)”, “famous” as “in”, “famous” as “in”, “ "Pronouns", "to be loved by", "to be known as", "to and called", "alias", "alias", "nickname", " Abbreviations ~, "common names ~", "abbreviated ~" and the like are conceivable.
By counting only the words that appear with these phrases (words corresponding to the above-mentioned “˜”), it is possible to accurately narrow down related keywords that serve as aliases and facility images.

Furthermore, as a third method, the relevance of the related keyword with respect to the initial keyword can be calculated as “relevance”, and the related keyword can be selected based on the integrated value.
The degree of relevance is a numerical value reflecting “words associated with searching for a certain facility”, and a word having a high degree of relevance can be said to be a “keyword that is easy to speak during search”. This degree of association is a value calculated in consideration of the appearance with the above-mentioned specific wording in addition to the above-mentioned appearance frequency and various information described later, and determines the priority order of the extracted related keywords. It can be understood as a “weight coefficient”.
For example, since the storage capacity of the related keyword database 22 is limited, an infinite number of related keywords cannot be registered. Therefore, by adopting (storing) items with a high degree of relevance, it is possible to achieve both user convenience and storage capacity.

Specific examples of the calculation of the relevance level include: (1) A keyword that appears with a specific wording on the WEB information 30 is set to have a higher relevance level than the other keywords. (2) Appearance position on the WEB page The higher the degree of relevance, the higher the degree of relevance is set for the keywords located in the upper part of the page.
Furthermore, (4) when the target initial keyword itself exists in the text, the smaller the distance from the initial keyword, that is, the number of words or the number of characters included between the initial keyword and the target keyword, (5) A high degree of relevance may be set for keywords that are emphasized such as a large font size, bold font, underline, and character color different from the surroundings.

The formula for calculating the relevance level reflecting the above relevance level setting method is shown below.
When the count value of the keyword “KW” is represented by Count (KW) and the i-th keyword “KW” is represented by KW (i),
Count (KW) = Σ _{i = 1} ⁿ (1 + IF _α (KW (i)) * α + IF _β (KW (i)) * β + IF _γ (KW (i)) * γ +...)
It becomes.
However, IF _α (KW (i)), IF _β (KW (i)), IF _γ (KW (i))... Are set when the relevance of the keyword (KW (i)) is set high. The conditions are as follows.
If the condition for setting the relevance level is satisfied, the satisfied conditions (IF _α (KW (i)), IF _β (KW (i)), IF _γ (KW (i)) ...) is set to 1.
Therefore, when a certain condition is satisfied, bonuses α, β, γ... Are added to the count value (Count (KW)) of the keyword (KW).
On the other hand, when the keyword “KW” simply appears without satisfying any condition, all the bonus portions are 0, so 1 is calculated as the count value (Count (KW)).

By dividing the count value (Count (KW)) of the keyword “KW” thus obtained by the count values for all keywords (KW = 1 to n) as in the following equation, the relevance of the keyword “KW” is calculate.
Cor (KW) = Count (KW) / Σ _{KW = 1} ⁿ (Count (KW))
It is desirable to adjust the bonus α, β, γ... For each condition (condition for setting a high degree of association) by experimentally calculating the web text analysis.
The data of the related keyword database 22 in FIG. 4 and FIG. 5 to be described later shows an example extracted by the above degree of relevance.
In addition, when the voice search unit 17 recognizes the user's voice input through the microphone 10, the relevance degree is added to the relevance degree, and a related keyword having a high relevance degree is more easily recognized than a keyword having a low relevance degree. You can also use it.

Next, the timing at which the related vocabulary collection unit 21 collects related keywords will be described.
In the present embodiment, the related vocabulary collection unit 21 is arranged in the information center 2 and is separated from the in-vehicle navigation device 1, so that it is possible to always execute related keyword collection processing, Any timing is possible.

As an example, the related vocabulary collection unit 21 periodically executes and monitors whether there is no change in the search result of the WEB information described above and whether there is a change in the text information of the WEB page, for example, once a day. You may make it do. If there is a change, text analysis is performed only for the changed WEB page.
The web page address and the page contents extracted in the previous search are cached, and the difference between the result of the new search and the previous search result is taken to perform analysis processing on the web page that has become the difference. It can be updated as a related keyword for new collection.

  Note that a large storage capacity is required to cache and compare the text content of the WEB page itself, but it is possible to use RSS (Rich Site Summary), which is an expression format of WEB page update information used on the Internet. For example, it is possible to know only the updated part of the WEB page.

  As another collection timing, on the premise of a sufficiently fast server function of the information center 2 and a sufficiently fast communication environment, the related vocabulary collection unit 21 performs the WEB when the dictionary update command is received from the navigation device 1 side. You may make it perform a search and perform a related keyword extraction process, and return the result to the navigation apparatus 1 side.

Next, a dictionary update method in the dictionary update unit 18 of the navigation device 1 will be described.
Here, since the information to be updated is transferred via the communication units 14 and 20, the difference is updated from the viewpoint of reducing the traffic.
In the information stored in the related keyword database 22 shown in FIG. 4, time stamp information is recorded as a version ID for each related keyword, and the current dictionary version ID is recorded in the recognition dictionary database 13 shown in FIG. Therefore, only the difference can be communicated by comparing the versions of the related keywords.

For this reason, the dictionary update unit 18 transmits the dictionary version ID of the recognition dictionary database 13 to the related vocabulary collection unit 21 of the information center 2 together with a dictionary update command.
The related vocabulary collection unit 21 refers to the related keyword database 22, compares the received dictionary version ID of the recognition dictionary database 13 with the version ID of each related keyword in the related keyword database 22, and recognizes in the related keyword database 22. A related keyword that is newly updated from the dictionary version ID of the dictionary database 13 is extracted.

In the example of the recognition dictionary database 13 shown in FIG. 2, the dictionary version ID is 20071201000000. Therefore, the related keyword database whose version ID is hatched with a diagonal line of 20071201000000 or more is extracted from the related keyword database 22 shown in FIG. .
When the related keyword as a target is extracted, the related vocabulary collection unit 21 transmits the related keyword as related vocabulary update information to the dictionary update unit 18 of the navigation device 1.
The dictionary update unit 18 registers the received related keywords in the recognition dictionary database 13.

As a result, as shown in FIG. 5, for example, the dictionary version ID becomes the latest, and a new related keyword (related keyword shown in bold in FIG. 5) is added.
An example of a language model using the added related keyword is shown in FIG. In the language model of FIG. 3, new related keywords (Ajisai Temple, Yokohama Marinos ...) are added to the registered vocabulary of the related keywords that are NULL, and the voice search unit 17 uses the keywords as recognition vocabularies. Search becomes possible.

It should be noted that the frequency or timing of updating the dictionary, but because there are a myriad of facilities nationwide, in order to obtain related keywords for all facilities, storage capacity, communication speed, data update processing speed, etc. The condition must be achieved in a high dimension.
However, in the case of use with a car navigation device, it is not highly likely that a facility in the whole country is set as a destination by a specific user.
Also, the analyzed related keywords need not be updated many times in one day.
Therefore, the area to be searched can be limited to some extent depending on the area where the user lives and the personality of the user, and the frequency of updating is considered to be sufficient for each trip in which the navigation device 1 is used.

Therefore, the time when the ignition switch is turned on for starting the engine of the vehicle and the navigation device 1 is activated is set as a normal update timing.
Since the information center 2 is always operating, the activation of the navigation device 1 is activated as a navigation system.
In the case where the ignition switch is repeatedly turned on and off a plurality of times within one day, it may not be updated if the dates are the same.

Further, the update area is limited to the area where the user lives or is frequently used as a part of the area of the related keyword currently stored in the recognition dictionary database 13, and the time when the area is exceeded is set as the update timing. .
For example, as shown in FIG. 7, the point where the engine is started and the navigation device 1 is started (or the point registered as home) is P00, the facility searched in the past or the registered point is P01, P02, When it is assumed that it is P03, at the activation point P00, the areas at a predetermined distance from the points P00 to P03 are set as update ranges A00, A01, A02, and A03.
Furthermore, the routes from the point P00 to the points P01, P02, and P03 may be R01, R02, and R03, and this region may be the update range.

Thus, when the navigation device is activated, the dictionary update unit 18 communicates with the related vocabulary collection unit 21 of the information center 2 to set the update ranges A00, A01, A02, A03 and the routes R01, R02, R03 as the target range. The related keywords in the recognition dictionary database 13 are updated.
In the example of FIG. 7, a predetermined distance area from each point P00 to P03, that is, a circular area is set as the update range A00 to A03, but an administrative area such as the prefecture to which each point belongs or its neighboring prefecture is also set as the update range. Also good.
In addition, even if a point is not registered, it is possible to analyze the use area based on the travel history and set the area as the update target area.

Next, with regard to dictionary update, specific processing between the dictionary update unit 18 and the related vocabulary collection unit 21 will be described with reference to the flowchart of FIG.
Here, it is assumed that the related keyword is updated based on the current position, the registered position, and the history position information.
As described above, this update process is started when the ignition switch is turned on and the navigation device 1 is activated.
First, in step 100, the dictionary update unit 18 acquires the current position of the host vehicle from the current position acquisition unit 50.

In step 101, the dictionary update unit 18 checks whether or not dictionary update has already been performed after activation. This can be determined by using a flag indicating whether or not the dictionary is updated. If this is the first dictionary update process, the process proceeds to step 103. If the dictionary update process has already been performed once, the process proceeds to step 102.
In step 103, the dictionary update unit 18 acquires the registered position and destination history position information from the navigation function unit 19.
In step 104, the dictionary updating unit 18 refers to the use frequency for each region recorded by the voice search unit 17 based on the facility search use history with respect to the acquired registration position and history position, and the use frequency is predetermined. Focus on more than the value.

In step 105, the dictionary update unit 18 acquires the dictionary version ID recorded in the recognition dictionary database 13.
Thereafter, in step 106, the dictionary updating unit 18 stores the acquired current position, the dictionary version ID of the recognition dictionary database 13, the registered position narrowed down to a frequency of use equal to or higher than a predetermined value, and history position information. It transmits to the related vocabulary collection part 21 of the center 2.

In step 107, the related vocabulary collection unit 21 selects a data range to be updated with reference to the related keyword database 22 based on the current position, registered position, and history position information received from the dictionary update unit 18.
For example, if the initial keyword such as the facility name, latitude / longitude information, and prefecture information included in the transmitted information is included in the information stored in the related keyword database 22, the information including the initial keyword is displayed. Select as update target data range.
In step 108, the related vocabulary collection unit 21 reads the version ID of the related keyword corresponding to the update target data range selected in step 107.

In step 109, the related vocabulary collection unit 21 compares the version ID of the related keyword with the dictionary version ID of the received recognition dictionary database 13 and determines whether difference data exists.
Specifically, if there is a related keyword version ID newer than the dictionary version ID in the recognition dictionary database 13, the process proceeds to step 110, and if not, the process returns to step 100.
In step 110, the related vocabulary collection unit 21 extracts a related keyword having a version ID newer than the dictionary version ID in the recognition dictionary database 13 as a related keyword to be updated.

In step 111, the related vocabulary collection unit 21 transmits the related keyword extracted in step 110 to the dictionary update unit 18 of the navigation device 1.
In step 112, the dictionary update unit 18 adds the related keyword to be updated received from the related vocabulary collection unit 21 to the recognition dictionary database 13 as a new related keyword. At the same time, the dictionary version ID of the recognition dictionary database 13 is updated with the latest version ID of the received related keyword.
Here, the navigation device 1 is in a standby state for voice search by the voice search unit 17.
After step 112, the process returns to step 100.

In the next flow, since dictionary update has already been executed, the process proceeds from step 101 to step 102.
In step 102, the dictionary updating unit 18 checks whether or not the current position of the host vehicle acquired in step 100 exceeds a predetermined area set in advance.
When the current position of the host vehicle does not exceed the predetermined area, the process returns to step 100, and when the current position exceeds the predetermined area, the process proceeds to step 103.
According to the above flow, when the navigation device 1 is activated, the related keyword in the recognition dictionary database 13 is first updated only once, and then when the current position of the host vehicle exceeds a predetermined area while the navigation device 1 is in operation. The related keyword is updated again.

Note that, as related keywords are updated and increased, there is a possibility that a plurality of initial keywords, that is, facility names, correspond to a certain related keyword, that is, so-called related keywords overlap.
In this case, when searching the database with the recognized related keyword, the voice search unit 17 presents a duplicate initial keyword, that is, a plurality of facility names on the speaker display 16 and can be selected by the user interactively. Like that.

  In the present embodiment, the microphone 10 and the AD conversion unit 11 constitute the voice input unit in the invention, the navigation function unit 19 corresponds to the guidance information generating means, and the DA conversion unit 15 and the speaker display 16 constitute the output unit. is doing.

The present embodiment is configured as described above, and in the navigation system that performs voice recognition and provides information requested by the user, the related vocabulary collection unit 21 that collects related keywords related to the initial keyword from the WEB information 30, and the collection thereof Since the configuration includes the dictionary update unit 18 that reflects the result in the recognition dictionary database 13, the voice search unit 17 can perform a voice search using the initial keyword and the related keyword stored in the recognition dictionary database 13, Even a user who does not know the official vocabulary can obtain a desired search result.
In addition, it is not necessary to register related keywords in advance, and a recognition dictionary database can be automatically generated and updated for what is considered to be highly necessary for the user.
When the initial keyword is the name of the facility, the user can search for the desired facility even when the user does not know the official name, which is particularly effective for route guidance to the facility.

  In particular, when the related vocabulary collection unit 21 collects related keywords, it extracts a keyword having an appearance frequency of a predetermined value or higher in a WEB search, or extracts a keyword that appears with a specific wording, thereby providing a highly reliable relationship. Keywords can be collected.

  In addition, when there is another facility name B including the facility name A or an area name C including the facility name A with respect to the facility name A of the initial keyword, the facility name B or the area name C is changed. The WEB page searched using is excluded from the analysis target, so the area with the same name as the facility name, such as the WEB page of another facility with a similar facility name or “Kokubunji City” for a temple and shrine named “Kokubunji”, for example If the WEB page exists, pages other than the WEB page of the original facility are excluded from analysis, and related keywords with higher reliability can be collected.

  In addition, since the page containing the facility type name other than the facility type corresponding to the facility name of the initial keyword is excluded from the analysis target, even if there is a facility name of the same name with different facility types, WEB pages other than the corresponding facility type are excluded from analysis, and reliability can be improved.

Further, when the related vocabulary collecting unit 21 collects related keywords, the related vocabulary collecting unit 21 collects related keywords with higher reliability by using the degree of relevance calculated by weighting based on the display position and the display format. Can do.
More specifically, the weighting value at the time of calculating the relevance is included in the WEB page by increasing the display position of the appearing vocabulary as it is closer to the title, heading, and page top in the page. Important information can be preferentially extracted from the information to be displayed, and the reliability of related keywords can be improved.

In addition, regarding the weighting value at the time of calculating the relevance level, the display position of the appearing vocabulary is set to be larger as the distance from the appearance position of the initial keyword is smaller, so that the related keyword closer to the initial keyword is preferentially collected. , Reliability of related keywords can be improved.
A related keyword that is close to the initial keyword in the text information by using at least one of the number of characters, number of morphemes, number of phrases, number of sentences, number of paragraphs included from the initial keyword to the target vocabulary as the distance from the appearance position As a result, it is possible to improve the reliability of related keywords.

  In addition, when the format of the vocabulary that appears, that is, the font size, the presence or absence of underline, the typeface, etc., is a different emphasis format than usual, by setting a large weight, the emphasized related keywords are given priority in the text information. Can be collected and the reliability of related keywords can be improved.

In addition, as a weighting of other relevance, by setting a large weighting for the vocabulary that appeared with a specific wording, it is possible to efficiently collect related keywords that are typical expressions that appear with a specific wording. be able to.
In particular, as a specific wording, “speaking of”, “to (famous) (as) famous”, etc. are expressions typically used to describe alternative names of initial keywords such as facility names. Because of the format, it is possible to efficiently collect related keywords by increasing the weight of the vocabulary appearing as such fixed phrases.

  The dictionary updating unit 18 updates the recognition dictionary database 13 when the navigation device 1 is started or when the current location of the vehicle on which the navigation device 1 is mounted exceeds a predetermined area. Thus, the latest related keywords can be added to the recognition dictionary database 13 while suppressing an unnecessary increase in the number of updates.

In addition, the recognition dictionary database 13 is updated by comparing the related keywords stored in the recognition dictionary database 13 with the related keywords collected by the related vocabulary collecting unit 21, and when there is a change, the related words collected by the related vocabulary collecting unit 21. Since the recognition dictionary database 13 is updated using keywords, it is sufficient to update only when there is the latest related keyword.
In particular, the dictionary updating unit 18 transmits the dictionary version ID of the recognition dictionary database 13 to the related vocabulary collecting unit 21, and the related vocabulary collecting unit 21 receives the received dictionary version ID and the related keyword version collected by the related vocabulary collecting unit 21. Since the IDs are compared, the related keyword of the version ID newer than the dictionary version ID is extracted and transmitted to the dictionary update unit 18, the communication amount between the dictionary update unit 18 and the related vocabulary collection unit 21 can also be reduced.

  In addition, related keywords to be compared for updating are recognized as limited to related keywords in areas where the user is likely to search by using facility names that are within a predetermined range from registered or previously searched points. The dictionary database 13 can be updated, and a voice search using the latest related keywords can be provided at a low update cost.

  In addition, since the voice search unit 17 monitors the user's facility search use history and records the use frequency for each region, the dictionary update unit 18 updates the related keyword only for the region where the use frequency is high. Voice search using the latest related keywords can be provided according to the actual use of facility search.

  In addition, when there is an initial keyword having the same related keyword at the time of the voice search by the voice search unit 17, the plurality of initial keywords are selected and presented, so that the user has a plurality of results of searching with the related keyword. Even in this case, a desired search result can be obtained by selecting from the presented options.

  In addition, by placing the related vocabulary collection unit 21 in the information center 2 separated from the navigation device 1, the related vocabulary collection unit 21 collects and holds the latest related keywords at any time, such as at all times, while the navigation device 1 of the client. Since the recognition dictionary database 13 is updated only when the use is started or necessary, the latest recognition dictionary database for voice search can be held at a low update cost.

  In the embodiment, the update process is started by a request from the navigation device 1 side. For example, the update information is included in the broadcast wave from the related vocabulary collection unit 21 on the information center 2 side, for example, at an arbitrary timing. It is good also as a structure which transmits the update information of a related keyword to the navigation apparatus 1. FIG.

Next, a modified example will be described.
In the above embodiment, the client-side navigation device 1 has a voice search function, and the information center 2 has a related keyword collection and storage process. Is integrated into the navigation device to make it stand-alone type.
FIG. 9 shows an overall configuration of a navigation device according to a modification.
In the embodiment of FIG. 1, the related keyword database 22 is deleted, and the navigation device 1A has the related vocabulary collection unit 21, so that the navigation device 1A constitutes a navigation system.

When updating the dictionary, the related vocabulary collection unit 21 of the navigation device 1A itself connects to the WEB information 30 of the Internet 3 to collect related keywords, and uses the result to update the direct recognition dictionary database 13 to perform voice recognition. Configured.
Since communication between the navigation device and the information center is also unnecessary, the communication units 14 and 20 are also deleted.
Other configurations are the same as those in the embodiment, and the same configurations as those in FIG. Transmission in steps 106 and 111 in the flowchart of FIG. 8 is read as direct transmission not via the communication units 14 and 20.

Also in this modified example, as in the embodiment, a voice search using the initial keyword and its related keyword stored in the recognition dictionary database 13 by the voice search unit 17 becomes possible, and a user who does not know a formal vocabulary can be used. However, a desired search result can be obtained.
In addition, the effects described in the paragraphs [0059] to [0071] are obtained as in the example.
Moreover, since this modification is a stand-alone type configuration, it has the advantage of being freed from restrictions associated with communication of collected related keywords and the like.

  In the flowchart of FIG. 8 in the above embodiment and the modified example, a predetermined area in which the dictionary update timing is set to a part of the area range of the related keywords held in the recognition dictionary database 13 when the navigation apparatus 1 is activated. As the other update timing, when the recognition dictionary database 13 exceeds the region of the related keyword currently held, when the voice search is started, the result of the voice search If there is no candidate with a sufficient recognition score (numbers such as likelihood, log likelihood, posterior probability, reliability, etc., used in general speech recognition), the result of the voice search was corrected or denied by the user When there is an explicit update request from the user, it can be set as appropriate, such as when a predetermined time has elapsed (periodic update).

  In particular, if the navigation device can download traffic information, difference map information, various Internet information, etc. as appropriate, and if it can upload the traveling history of the vehicle on a regular basis, the network connection timing should be set. If, for example, a surplus of traffic is used for updating the recognition dictionary database, information in the recognition dictionary database 13 can be more efficiently kept up-to-date.

In the embodiment and the modification, the update of the initial keyword itself is not described, but the initial keyword can also be updated according to the map information update method of a general navigation device.
In the navigation device, the point data, the facility data, and the road data are differentially updated. However, the differential dictionary database can be similarly differentially updated.
As a result, new facilities and the like can be searched by voice, and the number of facilities that the navigation apparatus can guide increases.

In the above-described embodiments and modifications, the dictionary update unit 18 performs a related keyword update process for facilities existing in the use area specified by the current position, registered position, history position information, and the like. Even if is updated, if there is an area that is not used for a predetermined period with reference to the facility search usage history monitored by the voice search unit 17, the dictionary update unit 18 preferably initializes the relevant keyword. .
As a result, related keywords in regions that are not used can be excluded from the search target, so that the recognition dictionary database 13 is updated in accordance with changes in the actual usage, the recognition accuracy by the voice search unit 17 is improved, and the keyword It is possible to save the memory area required for verification.

In the related keyword collection processing in the related vocabulary collection unit 21 of the embodiment and the modified example, the text included in the WEB page at the top of the search result is set as an analysis target. When searching a WEB page on an Internet search site, a search is often performed by combining a plurality of search keys. Therefore, information on a user's search keyword when searching a WEB page may be analyzed.
For example, if there is a famous “A Park” in Hydrangea and you search for “A Park Ajisai” as a search key, this information will be recognized by the user as “Related to A Park” and “Ajisai” as related keywords. It can be said that the information is very suitable for extracting related keywords.

Therefore, the search key (co-occurrence keyword) used together with the initial keyword (in this case, “A park”) is accumulated, and the higher frequency is extracted as the related keyword. The frequency value of the extraction criterion is set as a percentage according to the number of search keys.
Thus, when a user tries to check a specific facility, a keyword that is input at the same time can be used for a voice search, so that a voice search function having high affinity for the user can be provided. it can.

As a third method of the related keyword collecting process in the related vocabulary collecting unit 21, a WEB search is executed using the initial keyword as a search key as in the first embodiment, and the upper predetermined number of WEB pages in the search result are searched. The search key used when arriving from the Internet search site in accordance with the second method may be stored, and the frequently used search key may be used as the related keyword for the initial keyword.
In the second method described above, the keyword used together with the initial keyword is a candidate for the related keyword. However, in the third method, even a single search key is used to reach the WEB page. All search keys are candidates for related keywords.

For example, if the search key used to reach the “A park” page is “hydrangea”, this is a related keyword candidate. It is possible to analyze the frequency of search keys that are candidates and to extract the higher ranks as related keywords.
As a result, since the keyword used in the search engine when the user reaches the WEB page of the initial keyword, that is, the WEB page of the desired facility can be used for the voice search, the voice having high affinity for the user. A search function can be provided.

  In the above embodiment and modification, when updating the related keyword in the recognition dictionary database 13, the difference is updated from the viewpoint of processing amount and communication amount reduction. Of course, the data may be overwritten and updated.

  In the embodiment and the modification, the update of the recognition dictionary database in the navigation device has been described. However, the present invention is not limited to the navigation device, and the present invention is, for example, facility guidance using a telephone line or recognition for voice search in the WEB. It can also be applied to updating a dictionary database.

It is a block diagram which shows the whole structure of an Example. It is a figure which shows the example of a recognition dictionary database. It is a figure which shows the example of a language model. It is a figure which shows the example of a related keyword database. It is a figure which shows the example of a recognition dictionary database. It is a figure which shows the example of a language model. It is a figure which shows the example of the dictionary update object range. It is a figure which shows the flow of the dictionary update process in an Example. It is a block diagram which shows the whole structure of a modification.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1, 1A navigation apparatus 2 Information center 3 Internet 5 Navigation system 10 Microphone 11 AD conversion part 12 Calculation part 13 Recognition dictionary database 14, 20 Communication part 15 DA conversion part 16 Speaker display 17 Voice search part 18 Dictionary update part 19 Navigation function Part (guidance information generating means)
21 Related Vocabulary Collection Unit 22 Related Keyword Database 30 WEB Information 50 Current Location Acquisition Unit

Claims (26)

An audio input unit for acquiring an audio signal;
A recognition dictionary database in which initial keywords are stored in advance;
A voice search unit for recognizing a keyword stored in the recognition dictionary database from a voice signal acquired by the voice input unit;
Guidance information generating means for generating guidance information based on the recognition result by the voice search unit;
In an information guidance system comprising an output unit for presenting guidance information,
A related vocabulary collection unit that collects vocabulary having a semantic relevance to the initial keyword from WEB information as a related keyword;
A dictionary updating unit that adds the related keywords collected by the related vocabulary collecting unit to the recognition dictionary database at a predetermined timing;
The information retrieval system, wherein the voice search unit uses the initial keyword and the related keyword as keywords to be recognized. The initial keyword is a facility name capable of route guidance by navigation,
The information guidance system according to claim 1, wherein the guidance function is route guidance to a predetermined point specified by the facility name. The related vocabulary collection unit performs a WEB search using the initial keyword, analyzes text information included in the searched WEB page, and collects vocabulary having an appearance frequency of a predetermined value or more as the related keyword. The information guidance system according to claim 1 or 2, characterized by the above-mentioned. The related vocabulary collection unit performs a WEB search using the initial keyword, analyzes text information included in the searched WEB page, and expresses an alias for the initial keyword or an event representing the initial keyword. 3. The information guidance system according to claim 1, wherein vocabulary having an appearance frequency equal to or higher than a predetermined value is collected as the related keyword with a specific wording. In addition to the appearance frequency, the related vocabulary collection unit further calculates a relevance level weighted for each vocabulary, and extracts the related keywords by narrowing down to a vocabulary whose integrated value of the relevance level is a predetermined value or more. The information guidance system according to claim 3 or 4, characterized by the above-mentioned. 6. The information guidance system according to claim 5, wherein the weighting for calculating the relevance is set for each vocabulary based on the display position and display format. 7. The information guidance system according to claim 6, wherein the weight for calculating the relevance is set to a larger value as the display position of the vocabulary is closer to the title, headline, and page head in the WEB page. 8. The information guidance system according to claim 6, wherein the weight for calculating the relevance is set such that the display position of the vocabulary is larger as the distance from the appearance position of the initial keyword is closer. 9. The information according to claim 8, wherein the distance from the appearance position is measured by at least one of the number of characters, the number of morphemes, the number of phrases, the number of sentences, and the number of paragraphs included from the initial keyword to the target vocabulary. Guidance system. The weight for calculating the relevance is set to a value larger than the normal format when the format represented by any of the vocabulary character size, the presence or absence of the underline, and the typeface is different from the most commonly used format in the WEB page. The information guidance system according to any one of claims 6 to 9, wherein 6. The weighting for calculating the degree of association is set to a large value for a vocabulary that appears with a specific wording for expressing an alias for the initial keyword or an event representative of the initial keyword. The information guidance system described. The specific phrases include at least "speaking of", "famous for", "famous for", "famous as", "known for", "known for", " Known as "," de famous "," ~ is famous "," as famous "," as famous "," as famous "," as famous "," no pronoun "," as " `` Preferred '', `` Preferred as '', `` Called '', `` Also called '', `` Alias- '', `` Alias- '', `` Nickname- '', `` Abbreviation- '', `` Common name- '', and `` 12. The information guidance system according to claim 4 or 11, wherein any one of "abbreviated" is included. When the related vocabulary collecting unit collects the related keywords, if there is another facility name B including the facility name A as the initial keyword or a region name C including the facility name A, the facility 13. The information guidance system according to claim 2, wherein a WEB page searched by executing a WEB search using the name B or the area name C is excluded from the analysis target. The recognition dictionary database stores the facility name as the initial keyword in association with the facility type,
The related vocabulary collecting unit, when collecting the related keywords, excludes a WEB page in which a vocabulary of a facility type name other than the facility type corresponding to the facility name is included at a frequency of a predetermined value or more from an analysis target. 14. The information guidance system according to any one of claims 2 to 13, characterized in that The related vocabulary collection unit accumulates vocabulary inputted together with the initial keyword as a co-occurrence keyword in a search service on the Internet, and extracts a vocabulary having a high frequency of the co-occurrence keyword as a related keyword for the initial keyword. The information guidance system according to claim 1 or 2, characterized by the above-mentioned. The related vocabulary collecting unit accumulates a search keyword when a predetermined number of WEB pages searched using the initial keyword are reached from an Internet search site, and the frequency of the search keyword is high. The information guidance system according to claim 1 or 2, wherein a vocabulary is extracted as a related keyword for the initial keyword. The dictionary updating unit requests the related vocabulary collecting unit to collect related keywords for the initial keyword at the time of activation or when the current location exceeds a predetermined area, and the related vocabulary collecting unit collects and returns the related keywords The information guidance system according to any one of claims 1 to 16, wherein the recognition dictionary database is updated using a keyword. The related vocabulary collection unit, when there is a difference between the related keyword stored in the recognition dictionary database and the related keyword collected by the related vocabulary collection unit, The information guidance system according to claim 17, wherein the information is returned to the dictionary update unit. The voice search unit has a function of monitoring a facility search use history and recording a use frequency for each region,
The information dictionary system according to claim 17 or 18, wherein the dictionary updating unit updates the recognition dictionary database for an area where the use frequency is a predetermined value or more. The voice search unit has a function of monitoring a facility search use history and recording a use frequency for each region,
The information guidance system according to any one of claims 17 to 19, wherein the dictionary update unit deletes the related keyword in the recognition dictionary database for an area where the use frequency is lower than a predetermined value. 21. The voice search unit according to any one of claims 1 to 20, wherein when there are a plurality of overlapping initial keywords for the recognized related keyword, the voice search unit presents the plurality of initial keywords to the user through the output unit in a selectable manner. The information guidance system according to claim 1. The recognition dictionary database, the voice input unit, the voice search unit, the dictionary update unit, and the output unit are arranged on the client side, and the related vocabulary collection unit is arranged on the server side, which is configured as a client / server type. The information guidance system according to any one of claims 1 to 21. 22. The recognition dictionary database, a voice input unit, a voice search unit, a dictionary update unit, a related vocabulary collection unit, and an output unit are integrated to form a stand-alone type. Information guidance system. In an information guidance system for recognizing a keyword stored in a recognition dictionary database by voice recognition and providing guidance information,
Relevant keywords that have a semantic relevance to the initial keywords stored in advance in the recognition dictionary database are collected from the WEB information,
A recognition dictionary database update method, comprising: adding the collected related keywords to the recognition dictionary database to update the keywords. The initial keyword is a facility name,
The recognition dictionary database is arranged on the client side,
The related keyword collection is performed at an arbitrary timing on the server side,
25. The recognition dictionary database update method according to claim 24, wherein the addition of the collected related keywords to the recognition dictionary database is executed when the client side is activated or when the current location of the client side exceeds a predetermined area. . 26. The recognition dictionary database update according to claim 24 or 25, wherein the collection of the related keywords is performed by analyzing text information included in a WEB page and extracting words having an appearance frequency equal to or higher than a predetermined value. Method.

Images