JPH10319990A - Compressing and developing methods of word dictionary data, voice recognizer, navigation system with voice recognition function and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To convert into the word dictionary data in the graphic form which is effective in word voice recognition and to conduct the data compression so as to effectively and easily develop during a transfer. SOLUTION: The graphic data shown in the diagram are converted into the compressed dictionary data by a second conversion process as 'ho,ah,,n,mi,,ma,cho,,chi,,W,1,,u,,J,8,W,2,ki,,ta,,W,3,,i,,o,,i,,W,4,'. The label assigned to a summit is rearranged using the base of proceeding order scanning of 'ho'→'n'→'ma'→'chi'→'W'.... Since 'ho' has a brother 'ah', 'n' has a brother 'mi' and 'ma' has a brother 'cho', these are also rearranged in accordance with 'a second scanning order' in which a scanning is also conducted. Note that '' is set as identification data to distinguish own brother relationship from other relationships.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for compressing word dictionary data, which is effective for voice recognition and the like used when a destination setting or the like in a navigation system can be input by voice. And a voice recognition device, a navigation system with a voice recognition function, and a recording medium.

[0002]

2. Description of the Related Art Conventionally, a method of recognizing an input voice by comparing it with a plurality of candidate patterns to be compared stored in advance and an apparatus using the method have been already put into practical use. I have. For example, a voice recognition device used for inputting a destination or the like by voice in a car navigation system or the like is such. When recognizing the word voice, first, an external input voice is analyzed to extract acoustic features, and the time-series data of the extracted acoustic features is converted to a word by a well-known DP matching method or the like. Using the acoustic features registered for each word to be recognized in the dictionary, the data sequence is divided into the data sequences most similar to the acoustic features of any of the words, and each of the divided data sequences is assigned a corresponding value. The word sequence of the input speech is recognized by assigning a word represented by the acoustic feature. Then, the recognition result is output to the navigation device.

Considering the nature of navigation, the location data at the center of this instruction data is the above-mentioned destination and the like. Since the place name data is a part of the map data used for the navigation processing, it is assumed that a lot of situations are newly added or updated. Therefore, in order to cope with the new addition or update, it is necessary to externally supply word dictionary data corresponding to the place name data to the speech recognition device.

However, when large-scale word dictionary data is transferred from the navigation device to the speech recognition device, the communication time is lengthened, which is not practical. In the case of a place name, there is a high possibility that a plurality of words having the same syllable data constituting the word exist. One of the reasons is that in addition to a specific place name, there are many place names such as north, south, east, west, and upper, middle, and lower. Also,
Another reason is that there may be more than one paraphrase when referring to the same location. That is,
"Honmachi" is referred to as "honmachi" or "honcho".

[0005] For this reason, in word dictionary data composed of a word group in which a plurality of words having the same syllable data exist, such as a place name, for example, instead of data created in a text format for each normal word, they are represented in a graph format. Is also advantageous. In other words, in the case of a tree structure, which is a typical example of the graph format, words having the same syllable data may be assigned to have the same parent.

However, a large amount of memory and a high-performance CPU are required to convert the data into the word dictionary data in the graph format.
Is required. Therefore, it is not practical to provide them on the side of the speech recognition device, and it is necessary to transfer word dictionary data created externally, which has the same problem as described above.

In order to solve such a problem, the present invention converts the data into word dictionary data in a graph format effective for use in, for example, word speech recognition, and further performs data compression so that the data can be expanded advantageously and easily at the time of transfer. An object of the present invention is to provide a method of compressing word dictionary data that can be applied, a method of expanding the word dictionary data, a speech recognition device, a navigation system with a speech recognition function, and a recording medium.

[0008]

Means for Solving the Problems and Effects of the Invention The word dictionary data conversion method according to the present invention provides a word dictionary data in which a word group in which a plurality of words having the same syllable data constituting a word exists exists in a text format. A method of compressing the syllable data in such a way that the text-form word dictionary data corresponds to a directed graph format similar to a tree structure, which is based on a tree structure but the number of paths reaching vertices is not necessarily one. And a first conversion process of assigning identification data indicating the end of a word to each vertex according to the first scanning order, and converting the data into graph dictionary data having data for identifying its own child and sibling for each vertex; With respect to the graph dictionary data obtained by the first conversion processing, identification data indicating syllables and word ends assigned to vertices of the directed graph similar to the tree structure is used. If there are siblings at the vertices while basically following the first scanning order, the siblings are rearranged according to a second scanning order in which the siblings are also scanned. A second conversion process of converting the word dictionary data into compressed dictionary data having identification data for discrimination, thereby substantially compressing the word dictionary data.

The method of the present invention is based on the premise that word dictionary data in which a word group in which a plurality of words having the same syllable data constituting a word exist in a text format is compressed. There are various word dictionaries having this "word group in which a plurality of words having the same syllable data constituting words" exist. For example, when a destination is input by voice in a navigation system as described later, Is a word group representing a place name. In the case of this place name, there is a high possibility that a plurality of words having the same syllable data exist. One of the reasons is that in addition to a specific place name, there are many place names such as north, south, east, west, and upper, middle, and lower. Another reason is that
It is also possible that multiple paraphrases may exist when referring to the same location. That is, "honmachi" is referred to as "honmachi" or "honcho".

Of course, the present invention is not limited to such a word group representing a place name, but may be applied to any other word group having similar properties, and the operation and effects described later can be suitably exerted. According to the method of the present invention, word dictionary data created in a text format is first converted into graph dictionary data by a first conversion process as described below, and then the graph dictionary data is converted by a second conversion process. The dictionary data is further converted to compressed dictionary data.

(1) First Conversion Processing The conversion into graph dictionary data is based on a tree structure, but the number of paths reaching the vertices is not necessarily one. Data and identification data indicating the end of a word are assigned to each vertex according to the first scanning order, and are converted into graph dictionary data in which each vertex has data for identifying its own child and sibling.

The "directed graph format similar to a tree structure" basically conforms to a tree structure, but the number of paths reaching the vertices is not necessarily one but may be plural. After that, they are allowed to "merge", so they will be called like this.

The syllable data assigned to each vertex and the identification data indicating the end of a word are provided with data for identifying their own children and siblings, so that they are arranged in the order of the parent and child in the tree structure. If each syllable data is connected, it becomes word data. In the case of Japanese, there are only about 110 syllable data, especially since there are many words in which the first one or two or more have the same syllable data, it is not necessary to have the same syllable data redundantly. For,
This is advantageous when storing as dictionary data. Of course, the syllable data itself is not extremely large in any language, not limited to Japanese, so the same syllable data does not need to be duplicated, which is advantageous when storing it as dictionary data. Some things are the same.

As the first scanning order, it is conceivable to adopt, for example, preorder traversal. Here, "scanning" refers to visiting each vertex in a certain order, but "preceding forward scanning" refers to
This refers to visiting the root, and then sequentially scanning the subtrees whose children are the roots (this scan is also a forward scan). Of course, other scans such as "inorder traversal" or "pos
Although there is a "torder traversal", it is possible to realize even with these scans.

(2) Second Conversion Processing In the second conversion processing, such graph dictionary data is further converted into compressed dictionary data. Compressed dictionary data is based on graph dictionary data, where syllables assigned to vertices of a directed graph similar to a tree structure and identification data indicating the end of a word are basically followed in the first scanning order, and siblings exist at vertices. Will also scan the brothers in the second
Are arranged in accordance with the scanning order, and identification data for distinguishing one's sibling relationship from others is provided.

That is, in the state of the graph dictionary data obtained by the first conversion processing, for each syllable data of each vertex,
It has data for identifying its own child and data for identifying its siblings. In the state of the compressed dictionary data obtained by this second conversion processing, the above-mentioned identification of its own child is performed. Data to perform the operation, the data amount is reduced by itself.

Furthermore, instead of having data for identifying siblings for each syllable data as described above,
Instead, they have identification data to distinguish their siblings from others. As the “identification data for distinguishing one's sibling relationship from others”, for example, data indicating that a predetermined syllable is the youngest sibling, or data indicating the number of siblings may be used. In this way, for example, syllable data having the same sibling relationship is 5
If one exists, the number of siblings is 5 for the five syllable data.
Or only the syllable data, which is the youngest sibling, may have data indicating that it is the youngest sibling.

In this specific example, in the state of the first graph dictionary data, data for identifying the sibling is required for every five syllable data, but in the state of the second graph dictionary data, the data of the sibling is required. Since it is sufficient to provide identification data only to the number “5” indicating the number or the youngest syllable data, the data amount is reduced. As a result, further data compression is realized.

As described above, there is no data for identifying one's own child. Further, as data for identifying siblings, "5" indicating the number of siblings for five syllable data is used. The reason why the identification data need only be given to the youngest syllable data is due to the following property of the second conversion processing. That is, the second conversion process scans the syllable data assigned to the vertices of the directed graph having a similar structure, basically, according to the first scanning order, if the vertices have siblings. They are rearranged according to the second scanning order. At the same time, it has identification data for distinguishing one's brother relationship from others. In other words, even if data for identifying parent-child relationships and sibling relationships is not set for each syllable data, the order of the re-arranged syllable data itself may indicate the relationship between parent-child and siblings in a tree-structured directed graph in some form. It will be shown by. However, since the order itself is not enough, identification data for distinguishing one's sibling relationship from others is provided. As a result, it is possible to first analyze the sibling relationship for each syllable data, and if the sibling relationship is known, by analyzing the sibling data together with the rearranged order of the syllable data, it is possible to determine the parent-child relationship that is not specified.

Accordingly, the amount of compressed dictionary data obtained in this way is relatively small in terms of data amount. For example, the communication time when externally transferring to a device using word dictionary data can be shortened. There are advantages. Further, in an apparatus using the word dictionary data, it is conceivable that the compressed dictionary data is expanded into graph dictionary data of a predetermined format for use, but the expansion processing can be easily performed.

On the other hand, as a method of expanding the compressed dictionary data compressed by the above-described word dictionary data compression method, for example, syllables in the compressed dictionary data and identification data indicating the end of words are arranged in the second scanning order. In addition, if there is a sibling at the vertex, and if there is identification data for distinguishing the sibling relationship from others, it is added to the identification data indicating the syllable and word end in the compressed dictionary data. And a method of setting data for identifying a child for each identification data indicating the syllable and the end of a word and developing the data in the second graph dictionary data.

In consideration of the conversion process in the second conversion process in the above-described word dictionary data compression method, on the contrary, each of the syllables in the compressed dictionary data and the identification data indicating the end of the word is converted into its own. The data to identify the child is restored. In other words, the identification data indicating the syllables and the end of the words in the compressed dictionary data is rearranged according to the second scanning order, and when there is a sibling at the vertex, the identification data for distinguishing the sibling relationship from the others. Since the data is set, the child can be easily analyzed from the parent-child relationship of the vertices of the directed graph having a similar tree structure, considering this point.

The second graph dictionary data expanded in this manner is not necessarily the same as the state of the graph dictionary data obtained by the first conversion processing by the above-described compression method. In the above-described graph dictionary data, the syllable data assigned to each vertex and the identification data indicating the end of the word have data for identifying their own children and siblings. In the graph dictionary data, it is not necessary to provide data for identifying one's brother for each syllable data assigned to each vertex and identification data indicating the end of a word. This information may be left as identification data for distinguishing the sibling relationship from others.

In any case, in addition to the syllable data assigned to each vertex and the identification data indicating the end of a word, data for identifying one's own child for each data, and for distinguishing a sibling relationship from others. , The second dictionary data can be sufficiently used as a word dictionary.

In the word dictionary data compression method described above, going through the first conversion process and the second conversion process is substantially data compression.
This is a special data compression based on the premise that syllable data constituting words are assigned to vertices of a directed graph similar to a tree structure. That is, it is not based on a general data compression technique such as variable-length coding, but on the property of syllable data of a word group to be compressed. Therefore, the word dictionary data converted into the compressed dictionary data by the second conversion process may be further compressed using a predetermined data compression method. The “predetermined data compression method” is a compression method using a general data compression technique such as the above-described variable-length coding.

In this way, further data compression can be realized. When the data is further compressed by the predetermined data compression method, the word dictionary data compressed by the predetermined compression method is also used as the expansion method of the compressed dictionary data. And the identification data indicating the syllable and the end of the word in the expanded compressed dictionary data is rearranged according to the second scanning order, and the vertex is expanded. If there is a sibling in the compressed dictionary data, the syllable and the end of the word are added to the identification data indicating the end of the syllable and the word based on the presence of identification data for distinguishing the sibling relationship from the others. It is conceivable that data for identifying one's own child is set for each piece of identification data shown, and is developed into second graph dictionary data.

A program for executing the above-described word dictionary data compression method and expansion method is provided, for example, as a program activated on the computer system side. In the case of such a program, for example, it is stored in a machine-readable recording medium such as a floppy disk, a magneto-optical disk, a CD-ROM, or a hard disk,
It can be used by loading it into a computer system and starting it as needed. Alternatively, the program may be stored in a ROM or a backup RAM as a machine-readable recording medium, and the ROM or the backup RAM may be incorporated in a computer system and used.

On the other hand, as a speech recognition apparatus using the word dictionary data compressed by the above-described word dictionary data compression method, for example, the following configuration can be mentioned. The configuration includes a word dictionary storage unit in which acoustic features of a plurality of words to be recognized are stored in advance for each word, and an acoustic analysis unit that extracts an acoustic feature by analyzing an externally input speech. And, time-series data of acoustic features extracted by the acoustic analysis unit, is divided into data sequences that are most similar to the acoustic features stored in the word dictionary storage unit,
A speech recognition unit for allocating a word represented by a corresponding acoustic feature quantity to each data sequence and recognizing a word sequence of the input speech, and an output unit for outputting a recognition result by the speech recognition unit to an external device. 6. A compressed dictionary data input, wherein said word dictionary data is input from an external device in a state converted into said compressed dictionary data by the word dictionary data compression method according to any one of claims 1 to 5. Means, and said compressed dictionary data input via said compressed dictionary data input means.
Compression dictionary data expansion means for expanding the second graph dictionary data by the word dictionary data expansion method described in 1., wherein the word dictionary storage means includes the second graph dictionary by the compression dictionary data expansion means It is characterized in that word dictionary data expanded in data is stored.

In the speech recognition apparatus, the acoustic feature of a plurality of words to be recognized is stored in the word dictionary storage for each word as the acoustic feature of the word to be recognized by the speech recognizer. ing. When a sound is input from the outside, first, the sound analysis unit analyzes the input sound and extracts an acoustic feature. Then, the voice recognition means
The time-series data of the acoustic features extracted by the acoustic analysis unit is divided into data sequences that are most similar to the acoustic features of the words stored in the word dictionary storage unit. The word sequence in the input speech is recognized by allocating the word represented by the corresponding acoustic feature.
When the speech recognition means recognizes the word sequence in the input speech, the recognition result is output to the external device via the output means. In other words, only the necessary word series input by the user by voice is provided.

In this speech recognition apparatus, the word dictionary data is stored in the word dictionary storage means as follows. That is, the compressed dictionary data input means inputs the word dictionary data from the external device in a state where the word dictionary data is converted into the compressed dictionary data by the above-described word dictionary data compression method. Then, the compressed dictionary data expanding unit expands the input compressed dictionary data into the second graph dictionary data by the above-described word dictionary data expanding method. This second
Is stored in the word dictionary storage means.

As described above, since the word dictionary data can be input from the external device in a state converted to the compressed dictionary data, the communication time when inputting the word dictionary data from the external device can be reduced. Further, the word dictionary data stored in the word dictionary storage means is data in the form of a graph dictionary, which is an effective format for word speech recognition processing. For example, when input as text-format word dictionary data, it takes a considerable amount of time and memory to convert it into graph-format dictionary data, but such disadvantages do not occur.

In the above-described speech recognition device,
Although the word dictionary data is input from an external device in a state converted to the compressed dictionary data, the data is expanded into the second graph dictionary data and then stored in the word dictionary storage means.
The word dictionary storage means stores the compressed dictionary data input from an external device by the compressed dictionary data input means, and stores the compressed dictionary data read out from the compressed dictionary data storage means at power-on or during speech recognition processing. The second
A configuration that expands the data into the graph dictionary data can also be adopted.

In this way, the amount of word dictionary data that needs to be stored can be relatively small.
That is, if the word dictionary data storage means is constituted by, for example, a non-volatile memory, the capacity of the non-volatile memory may be small. If the expansion into the second graph dictionary data is performed during the execution of the speech recognition processing, the amount of volatile memory required only during the expansion is reduced.

As a navigation system including such a voice recognition device and a navigation device, for example, the following configuration can be considered. That is, the voice input means of the voice recognition device is used in order for the user to input an instruction of predetermined navigation-related data that needs to be specified when the navigation device performs the navigation process. Is used,
The said output means is the navigation system with a speech recognition function comprised so that the recognition result by the said speech recognition means may be output to the said navigation apparatus, The said navigation apparatus converts the said word dictionary data into said said 1 thru | or 4. A compressed dictionary data storage means for storing the converted dictionary data in a state converted by the method for compressing word dictionary data according to any one of 4. Dictionary data transfer means for reading dictionary data and transferring the dictionary data to the voice recognition device.

According to this navigation system, the compressed dictionary data storage means of the navigation device stores the word dictionary data in a state converted to the compressed dictionary data, and when the predetermined dictionary transfer is required, the compressed dictionary data storage means The compressed dictionary data is read and transferred to the speech recognition device.

In the case of the present navigation system, the navigation processing-related instruction data recognized by the voice recognition device is sent to the navigation device, and the navigation processing is performed based on the instruction data. Considering the nature of navigation, place name data forms the center of this instruction data. In addition to the unique property that there is a high possibility that there are a plurality of words having the same syllable data as described above, this place name data is newly added or updated due to the property that it is a part of the map data on the navigation device side. Many are assumed.

Therefore, to cope with the new addition or update, it is necessary to externally supply word dictionary data corresponding to the place name data to the speech recognition device. This is because it is conceivable that the addition or update of a place name is basically realized as data in a text format. However, in order to convert them into the above-mentioned word dictionary data in a graph format, a large amount of memory and a high-performance CPU are required. Needed. Therefore, it is not practical to provide them on the voice recognition device side, and it is necessary to supply word dictionary data in the form of a graph that is created in response to the addition or update of the data. However, even if the dictionary data is in the form of a graph, the data size is considerably large. Therefore, when transferring the word dictionary data from the navigation device to the speech recognition device, it is necessary to transfer the large-scale word dictionary data as it is. The communication time is long, which is still not practical. For this reason, the communication time can be reduced by transferring the compressed dictionary data to the voice recognition device after making it in the state of the compressed dictionary data.

Note that "when a predetermined dictionary transfer is required" may be, for example, when the map data is updated in the navigation device or when the battery is turned off. Also, the navigation device stores word dictionary data for transfer to the speech recognition device in the form of compressed dictionary data, but the conversion to the compressed dictionary data may be performed by the navigation device itself. Alternatively, the conversion may be performed by an external device, and the navigation device may input and store the compressed dictionary data from the external device. As described above, a large amount of memory and a considerably high-performance CPU are required to generate the word dictionary data in the graph format. Therefore, it is practical to convert the data by a work station or the like as an external device instead of providing them on the navigation device side.

[0039]

FIG. 1 is a block diagram showing the overall configuration of a car navigation system 2 as an embodiment of the present invention. The car navigation system 2 includes a position detector 4, a map data input device 6, an operation switch group 8, a control circuit 10 connected thereto, an external memory 12 connected to the control circuit 10, a display device 14, and a remote control sensor 15. And a voice recognition device 30. The control circuit 10 is configured as an ordinary computer, and includes a well-known CPU, ROM, RAM, I / O, and a bus line for connecting these components.

The position detector 4 is a well-known geomagnetic sensor 1.
6. GPS (Gl) for detecting the position of the vehicle based on radio waves from the gyroscope 18, the distance sensor 20, and satellites
It has a GPS receiver 22 for the obalPositioning System. These sensors and the like 16, 18, 20, 22
Since each has an error having a different property, it is configured to be used while being interpolated by a plurality of sensors. It should be noted that depending on the accuracy, a part of the above-described components may be used, and a rotation sensor for the steering wheel, a wheel sensor for each rolling wheel, or the like may be used.

The map data input device 6 is a device for inputting various data including so-called map matching data, map data and landmark data for improving the accuracy of position detection. As a medium, CD-R
Although OM is generally used, another medium such as a memory card may be used.

The display device 14 is a color display device. On the screen of the display device 14, the vehicle current position mark input from the position detector 4, the map data input from the map data input device 6, and the map data It is possible to superimpose and display additional data such as a guidance route displayed above and a mark of a set point described later.

The present car navigation system 2
Is input from the remote control sensor 15 via a remote control terminal (hereinafter referred to as a remote control) 15a or the position of the destination by the operation switches 8;
It also has a so-called route guidance function that automatically selects an optimal route from the current position to the destination and forms and displays a guidance route. As a technique for automatically setting the optimum route, a technique such as the Dijkstra method is known.
The operation switch group 8 is, for example, a touch switch integrated with the display device 14 or a mechanical switch, and is used for various inputs.

The voice recognition device 30 is used when the operation switch group 8 or the remote controller 15a is manually operated to indicate a destination or the like, and similarly, when the user inputs voice. This is a device that allows the user to specify a destination or the like.

The voice recognition device 30 includes a voice recognition unit 31
, A dialogue control unit 32, a voice synthesis unit 33, a voice input unit 34, a microphone 35 as "voice input means", a PT
A T (Push-To-Talk) switch 36 and a speaker 37 are provided. The voice recognition unit 31 performs a recognition process on the voice data input from the voice input unit 34 according to an instruction from the dialog control unit 32, and returns the recognition result to the dialog control unit 32. From the recognition result and the internal state managed by itself, the dialog control unit 32 instructs the speech synthesis unit 33 to issue a response voice to the voice synthesis unit 33 and to the control circuit 10 that executes processing of the system itself, for example, for navigation processing. Of the destination required by the user, and instructs to execute the setting process. Such processing is post-confirmation processing, and as a result, if this voice recognition device 30 is used, the destination of the navigation system can be indicated by voice input without manually operating the operation switch group 8 or the remote controller 15a. It becomes possible.

The voice input unit 34 converts the surrounding voice captured by the microphone 35 into digital data and outputs the digital data to the voice recognition unit 31. In this embodiment, the user operates the PTT switch 36. The voice is input via the microphone 35 while pressing. That is, when the PTT switch 36 is not pressed, the voice input unit 34 does not output the voice data to the voice recognition unit 31.

Here, the speech recognition unit 31 and the dialogue control unit 32
Will be described in more detail with reference to FIG.
As shown in FIG. 2, the voice recognition unit 31 includes a collation unit 31a and a dictionary unit 31b, and the interaction control unit 32 includes a post-processing unit 32a, a communication control unit 32b, and a dictionary control unit 32c. .

In the voice recognition section 31, a collation section 31a
Performs collation on the voice data acquired from the voice input unit 34 using the dictionary data stored in the dictionary unit 31b, and recognizes a word sequence in the input voice. Specifically, audio data input from the audio input unit 34 is sequentially subjected to acoustic analysis to extract an acoustic feature (for example, cepstrum), and to obtain acoustic feature time-series data obtained by the acoustic analysis. Then, the time-series data is divided into several sections by a well-known DP matching method, and processing as “speech recognition means” for determining which word each section corresponds to stored in the dictionary unit 31b is performed. Run.

The recognition result of the collating unit 31a is transmitted to the dialogue control unit 3
2 is sent to the post-processing unit 32a. The post-processing unit 32a transmits the data to the control circuit 10 via the communication control unit 32b, for example, to execute a post-determination process instructing to perform a predetermined process, or transmits the voice data to the voice synthesis unit 33. A process for instructing to sound is executed.

The dictionary control unit 32c of the dialog control unit 32
Expands data of “compressed dictionary data”, which will be described later, transferred from the control circuit 10 via the communication control unit 32b, further develops the dictionary, and writes the data to the dictionary unit 31b of the speech recognition unit 31. These “data expansion” and “dictionary expansion” will be described later.

Here, the operation of the car navigation system 2 will be briefly described. After the power of the car navigation system 2 is turned on, the driver can use the menu displayed on the display device 14 to display the guidance route using the remote controller 15a (the same operation can be performed with the operation switches 8; the same applies to the following description). If the user selects the route information display processing to display the information on the
When the same instruction is given from the dialogue control unit 32 to the control circuit 10 through the remote controller 15a by inputting a voice via the remote controller 15a, the following processing is performed.
That is, when the driver inputs a destination by voice or an operation of a remote controller or the like based on the map on the display device 14, the current location of the vehicle is obtained based on satellite data obtained from the GPS receiver 22, and the destination and the current location are determined. In between, the cost is calculated by the Dijkstra method, and the shortest route from the current position to the destination is determined as the guidance route. Then, the guidance route is displayed on the road map on the display device 14 so as to guide the driver to an appropriate route. The calculation process and the guidance process for obtaining such a guide route are generally well-known processes, and a description thereof will be omitted.

In the car navigation system 2, as described above, the voice recognition device 30 has the control circuit 10 transfer the compressed dictionary data, and the dictionary control unit 32c of the dialog control unit 32 transmits the compressed dictionary data. Is developed into a data dictionary and further developed into a dictionary, and then written into the dictionary unit 31b of the speech recognition unit 31. The compressed dictionary data is generated using an external device, for example, a workstation. In this procedure, as shown in FIG. 3, word dictionary data created in text format is converted into graph dictionary data by a first conversion process, and then the graph dictionary data is further converted by a second conversion process. Convert to compressed dictionary data. Furthermore, in the case of the present embodiment, general data compression such as variable-length coding is performed, and the compressed "compressed dictionary data" is written together with the map data and other data on the map CD for the navigation system. . Then, the data in the map CD is read into the control circuit 10 via the map data input device 6 (see FIG. 1), and is transferred to the voice recognition device 30 as described above.

The processing up to the conversion into the "compressed dictionary data" shown in FIG. 3E will be further described. First, “text dictionary data” shown in FIG. 3A will be described. As described above, this is word dictionary data created in a text format. In the case of the present embodiment, the word is mainly a place name because it is used for voice recognition of a destination in navigation. . Considering the nature of the word group representing the place name, there is a point that there are a plurality of words having the same syllable data constituting the word. One of the reasons is that, apart from a specific place name,
There are many places where east, west, north and south, upper middle, lower, etc. are added. Another reason is that a plurality of paraphrases may be present when indicating the same place. In other words, "honmachi" is replaced by "honmachi"
Or "honcho".

In order to facilitate understanding of the method of compressing word dictionary data in the present embodiment, the following description will be given using the following specific example as text dictionary data. (1) Book (town | cho) (2) Book (3) Book (4) Aioi In this specific example, numerals (1) to (4) indicate word numbers. () In (town | cho) indicates a priority order, and | indicates an option. In other words, (machi | cho) means selecting either "machi" or "cho", and (1) "hon (machi | cho)" replaces "honmachi" or "honcho". Show. However, “honmachi” has priority over “honcho”.
Although not used in this example, omission can be indicated by adding [].

Next, the processing of FIG. 3B for converting text dictionary data into graph dictionary data by the first conversion processing will be described. The first conversion process is described in an explanatory manner as follows: “Syllable data and word end are converted so as to correspond to a tree-structure-like directed graph format in which the number of paths reaching a vertex is not necessarily one based on a tree structure. The identification data shown is assigned to each vertex in the first scanning order, and is converted into graph dictionary data in which each vertex has its own child and sibling identification data. "

Since it is easier to understand by referring to specific examples, FIGS. 5 to 7 show specific examples of graph dictionary data as a result of the text dictionary data shown in FIG. 4 being converted by the first conversion processing. . 5 to 7 show the same graph dictionary data, FIG. 5 shows a case where the same graph dictionary data is expressed in a "directed graph format similar to a tree structure", FIG. 6 shows the same one in a pointer expression, and FIG. This is shown in an array expression.

The "directed graph format similar to the tree structure" shown in FIG. 5 basically conforms to the tree structure, but the number of paths reaching the vertices is not necessarily one but may be plural. Are allowed to "join" thereafter, so we will call it this way. The “first scanning order” in this case is a so-called “preceding forward scanning (preo scanning)”.
rder traversal). "Scanning" here
Refers to visiting each vertex in a certain order, while “pre-order scan” refers to visiting the root and then scanning a subtree with its children as roots. Scanning). Here, “parent” means the immediately preceding vertex, “child” means the next vertex, and “sibling” means vertices having the same parent.

That is, for example, the words (1) to (3) in the text dictionary data all have the same parent because they all begin with "ho". Also, since "honmachi" and "honcho" are the same up to "hon", the following "ma" and "cho" have a sibling relationship. Also, considering "hon (machi | cho)" and "homi", "n" and "mi" following "ho" have a sibling relationship. In FIG. 5, an arrow (→) indicates a side, a single circle (○) indicates a vertex, and a double circle (◎) indicates a received vertex, that is, a vertex for a word. The number at the reception vertex indicates the word number.

For example, since "honmachi" and "honcho" are words representing the same place name, they are finally merged to mean the same word. That is, the same reception vertex 1 is reached. Also, since “Homi” and “Homi Kita” have the same “Homi”, and since the word ends only with “Homi”, the accepted vertex 2 is shown after expressing “Homi”. , And thereafter "Kita", and the word ends there, indicating the accepted vertex 3.

On the other hand, FIG. 6 shows a child and a sibling for each label. Here, the syllable data "ho", "n",
Labels such as "ma" and "chi" are used, and "W" as data indicating that a word ends is also a label. FIG. 7 similarly shows a child and a sibling for each label, which is shown in an array expression. According to the first scanning order described above, that is, in this case, according to the pre-order scanning, the labels are “ho, ma, ma, w, w, cho, u, mi,
W, C, T, A, I, O, I, W ", and label numbers 1 to 17 are assigned respectively. And for each label, brother (Bros)
And child (Child) label numbers are set. For example, label number 13 is set for label "1" as a sibling and label number 2 is set as a child for label "ho". I don't know. " Note that the word number is set as child data for the label “W” as data indicating that the word has ended. For example, “W” at label number 5 indicates the end of the word “honmachi”, and since the word number is 1,
“1” is set as child data.

Next, "processing for converting graph dictionary data into compressed dictionary data by the second conversion processing" shown in FIG. 3D will be described. When the compressed dictionary data generated in the second conversion processing is described in an explanatory manner, “the first dictionary
The lexical data assigned to the vertices of the directed graph similar to the tree structure and the identification data indicating the end of words are compared with the graph dictionary data obtained in the conversion processing of the above, and siblings exist at the vertices while basically following the first scanning order. In this case, the siblings are rearranged according to the second scanning order in which the siblings are also scanned, and identification data for distinguishing one's sibling relationship from others is provided. "

Since it is easier to understand by referring to a specific example, a specific example of the compressed dictionary data obtained by converting the graph dictionary data shown in FIGS. 5 to 7 in the second conversion processing is shown in FIG. Shown in In other words, the compressed dictionary data in this case is
"Ho, a, ＼, n, mi, ＼, ma, cho, ＼, chi, ＼,
W, 1, ＼, ＼, J, 8, W, 2, ＼, ＼, ＼, ＼,
W, 3, ＼, ＼, お, ＼, ＼, ＼, W, 4, ＼ ”.

When the syllables assigned to the vertices of the directed graph similar to the tree structure shown in FIG. 5 and the identification data (label) indicating the end of the word are basically obeyed in the first scanning order, and there are siblings at the vertices Are sorted according to the "second scanning order" in which the siblings are also scanned. Since the above-described first scanning order is the preceding order scanning, “ho” →
Based on the pre-sequential scanning of "n" → "ma" → "chi" → "W" → ..., "ho" has a sibling "a" and "n" has a sibling "mi". Since "ma" has a sibling "cho", it scans them as well. In addition, “＼” is set as identification data for distinguishing one's brother relationship from others. For example, "@" is inserted after "ho, a".

The specific processing for converting the graph dictionary data shown in FIG. 7 to the compressed dictionary data shown in FIG. 8 will be described later, but the effect of such a conversion to the compressed dictionary data will be described. That is, in the state of the graph dictionary data (FIGS. 5 to 7) obtained by the first conversion processing, data for identifying its own child and data for identifying its sibling are provided for each syllable data of each vertex. However, in the state of the compressed dictionary data (FIG. 8) obtained by the second conversion process, the data does not have the data for identifying the above-mentioned own child. The amount is reduced.

Furthermore, instead of having data for identifying siblings for each syllable data as described above,
Instead, it has identification data (＼) to distinguish one's own siblings from others. In the case shown in FIG. 8, “＼” as the identification data is used as data indicating that the predetermined syllable is the youngest. That is, in the example shown in FIG. 8, since "a" is the youngest brother in the brother relationship "ho, a", "@" is set after that. In this example, there are only two siblings. For example, if there are five syllable data having the same sibling relationship, only the syllable data that is the youngest sibling to the five syllable data indicates that it is the youngest sibling. You just need to have the data. Therefore, in addition to the data amount reduction due to the absence of the data for identifying the own child described above, further data amount reduction can be realized. As a result, further data compression is realized.

The "identification data for distinguishing one's sibling relationship from the other" may be data indicating not only the above-mentioned younger brother but also the number of siblings, for example. That is, in the example of FIG. 8, if the number of siblings is found to be "2" for the sibling relationship "ho, a", that may be the case. Even in this case, data indicating the number of siblings may be provided regardless of the number of siblings, so there is no data indicating siblings for each syllable data as in the case of graph dictionary data. The amount will be reduced.

By the way, the compressed dictionary data does not have data for identifying its own child as described above. Further, as data for identifying siblings, the data of five syllable data are the youngest siblings. The reason why the identification data (＼) needs to be given only to the syllable data of the second conversion processing is due to the following property of the second conversion processing. That is, the second conversion processing is
The syllable data allocated to the vertices of the directed graph having a similar structure are rearranged according to a second scanning order in which, if there is a sibling at the vertex, the siblings are also scanned, basically following the first scanning order. ing. At the same time, it has identification data for distinguishing one's brother relationship from others. In other words, even if data for identifying parent-child relationships and sibling relationships is not set for each syllable data, the order of the re-arranged syllable data itself may indicate the relationship between parent-child and siblings in a tree-structured directed graph in some form. It will be shown by. However, since the order itself is not enough, identification data for distinguishing one's sibling relationship from others is provided. As a result, it is possible to first analyze the sibling relationship for each syllable data, and if the sibling relationship is known, by analyzing the sibling data together with the rearranged order of the syllable data, it is possible to determine the parent-child relationship that is not specified.

Therefore, the amount of compressed dictionary data obtained in this manner is relatively small in terms of data amount, and for example, the communication time when externally transferring to a device using word dictionary data can be shortened. There are advantages. Further, in an apparatus using the word dictionary data, it is conceivable to use the compressed dictionary data by expanding it into graph dictionary data of a predetermined format. However, the expansion processing can be easily performed.

Next, a description will be given of "specific processing for converting the graph dictionary data shown in FIG. 7 to the compressed dictionary data shown in FIG. 8" which will be described later in the above description. As shown in FIG. 10, when the process starts, a variable n is set to an initial value 1 in a first step S10, and Num (i) is set to an initial value 0 in a succeeding S20. Note that i = 0, 1,... N. After the completion of such initialization, the compression process starts in S30.

The details of this compression processing will be described with reference to FIGS. Note that FIGS. 11 and 12 are generalized as the compression (p) processing, and when p = 1, the compression (1) processing is performed.
Processing. Therefore, S30 in FIG. 10 is a compression (1) process. In the first step S110 in FIG.
= P is set. That is, in the compression (1) process, q =
p = 1. Then, in the subsequent S120, q and 0 are compared, and if q = 0, the process proceeds to S180.
If it is not 0, the process moves to S130.

In S130, Num (q) is set to n, and n is incremented (n = n + 1). Then, in S140, the contents of Label (q) are compared with "W" indicating the end of the word, and if the contents s of Label (q) is "W", the flow shifts to S170, and Label is changed.
If the content s of (q) is not “W”, the flow shifts to S150.

At S170, the contents s of Label (q)
Is “W”, so “W” is output and Chil is output.
Output d (q). On the other hand, in S150, Label
Since the content s of (q) is not “W”, Label (q)
Outputs itself. After the processing in S150 or S170, the process proceeds to S160.

In S160, the process returns to S120 with the variable q = Bros (q), and repeats the processing from S120.
On the other hand, in S180 to which the process proceeds when it is determined that q = 0 in S120, “＼” is output. And FIG.
The process proceeds to S190 as shown in FIG.

In S190, the first step S in FIG.
Similarly to 110, the variable is set to q = p. That is, in the compression (1) process, q = p = 1 also in this case. Then, in the subsequent S200, q is compared with 0. If q = 0, the main compression (p) process ends. If q = 0, the process proceeds to S210.

In S210, Label (q) is set to “W”.
If the content s of Label (q) is “W”, the process proceeds to S240, and the content s of Label (q) is “W”.
If not, the process proceeds to S220. In S220, Num
It is determined whether (Child (q)) is greater than 0. If Num (Child (q))> 0 (S22
0: YES), proceeding to S250, outputting "J" and outputting Num (Child (q)), and then returning to S2.
Move to 40. On the other hand, Num (Child (q)) ≦
If 0 (S220: NO), the process proceeds to S230,
A compression (Child (q)) process is performed. This compression (C
The hold (q)) process is a recursive read process, and performs the processes shown in FIGS. After the processing in S230, the process proceeds to S240.

If the content s of Label (q) is "W" in S210, and S230 or S250
In step S240, the process proceeds to the variable q = Bros
The process returns to S200 as (q) and repeats the processes from S200. The above is the description of the flowchart showing the compression processing. For easier understanding, the graph dictionary data shown in FIG. 7 is shown while being converted into the compressed dictionary data shown in FIG. explain.

FIG. 13 shows the compression (1) in S30 of FIG.
This is a conceptual illustration of the contents of the processing. In other words, while the compression (1) process is being executed, S230 in FIG.
Is performed (Child (q)). Since the compression (1) process is a process in the case of p = 1, S190
, Q = p = 1, and in S230, compression (Chi
ld (1)) processing. That is, referring to FIG.
Since hold (1) = 2, S230 in this case is a compression (2) process.

Thus, the compression (14)
When the processing ends, the compression (1) processing ends. In FIG. 13, point A indicates a time immediately before the shift to the compression (2) processing, point B indicates a time when the compression (2) processing is completed, and point C indicates a time when the compression (14) processing is completed. FIG. 14 shows the state (intermediate progress) at point A, point B, and point C.

First, the progress up to the point A will be examined in order. In the compression (1) process, p = 1, so that q = p = 1 in S110 of FIG. 11, and q = 1 in S120.
Since it is> 0, the process proceeds to S130. In S130,
Num (q (= 1)) = n (= 1) is set. This is because n = 1 is set in S10 of FIG. In S130, n is incremented, and
Let n = 2.

At S140, the contents of Label (1) are compared with "W" indicating the end of the word.
Since the content of (1) is “HO” and not “W”, S15
0, and outputs Label (1) itself, that is, “ho”. In subsequent S160, q = Bros (q) is set. However, since Bros (1) is 13 in FIG.
The process returns to S120 as 3.

At S120, which returns after the processing of S160, since q = 13 and not 0, the flow shifts to S130 to set Num (q (= 13)) = n (= 2).
Num (13) = 2 because n = 2 has been incremented in the previous processing of S130. Further, n is further incremented so that n = 3.

At S140, the contents of Label (13) are compared with "W" indicating the end of the word.
Since the content of (13) is “A” and not “W”, S1
The process proceeds to 50, and outputs Label (13) itself, that is, “a”. In S160, q = Bros
(Q), since Bros (13) is 0 from FIG. 7, the process returns to S120 with q = 0.

At S120, which returns after the processing at S160, since q = 0, the flow shifts to S180 to output “＼”. By the processing so far, “ho”, “a”, and “＼” are output, and as shown in the case of point A in FIG.
Num (1) = 1 corresponding to “HO” and Num (13) = 2 corresponding to “A” are set. Further, the compressed dictionary data in this state is "", "," as shown in the case of point A in FIG.

Next, the progress up to the point B will be examined in order. After the process proceeds to S180 described above and outputs “＼”, that is, S190 will be described. In S190,
As in the first step S110 of FIG. 11, the variable is set to q = p. That is, in this case, since the compression (1) process is still being performed, q = p = 1. And the following S20
Since q is not 0 at 0, the process shifts to S210 and La
Bel (1) is compared with “W”. Since Label (1) is “HO” and not “W”, the flow shifts to S220.

At S220, Num (Child
It is determined whether (1)) is greater than 0. Chil
Since d (1) is “2” from FIG. 7, Num (Chi)
ld (1)) = Num (2), and Num (2) = 0 as can be seen from the case where the state at the point A in FIG. 14 is shown. Therefore, a negative determination is made in S220, and the process proceeds to S230. In S230, a compression (Child (1)) process is performed. Since Child (1) = 2, this S
At 230, the compression (2) process is performed as a recursive read process.

In the compression (2) process, since p = 2,
In S110 of FIG. 11, q = p = 2, and in S120, since q = 2> 0, the process proceeds to S130. S1
At 30, Num (q (= 2)) = n (= 3) is set. In S130, n is further incremented to make n = 4.

At S140, the contents of Label (2) are compared with "W" indicating the end of the word. As can be seen from FIG. 7, the contents of Label (2) are "n" and "W".
Therefore, the flow shifts to S150, and outputs “n” of Label (2). In S160, q = Bros
(Q), since Bros (2) is 8 from FIG. 7, the process returns to S120 with q = 8.

At S120, which returns after the processing of S160, since q = 8 and not 0, the flow shifts to S130 to set Num (q (= 8)) = n (= 4). Num (8) = 4 because n = 4 has been incremented in the previous process of S130. Further, n is further incremented so that n = 5.

At S140, the contents of Label (8) are compared with "W" indicating the end of the word.
Since the content of (8) is “mi” and not “W”, S15
0, and outputs Label (8) itself, that is, “mi”. In subsequent S160, q = Bros (q) is set. However, since Bros (8) is 0 from FIG. 7, q = 0 and the process returns to S120.

At S120, which returns after the processing at S160, q = 0, so that the flow shifts to S180 to output “＼”. In the compression (2) process, “n”, “mi”, and “＼” are output by the processes so far.

Thereafter, during the above-mentioned compression (1) processing, S
As in the compression (2) process of 230, the compression (2)
In S230 during the processing, the compression (Child
(2)) Perform processing. As can be seen from FIG. 7, Child
Since (2) = 3, the compression (3) process is performed as a recursive read process in S230.

Although the details of the compression (3) processing are omitted, "ma", "cho" and "$" are output in the same manner as in the compression (2) processing described above. Then, in S230 during the compression (3) processing, compression (Child (3)) processing is performed. As can be seen from FIG. 7, Child (3) = 4, so in this S230, the compression (4) processing is performed as a recursive read processing.

In the compression (4) process, "chi" of Label (4) is output in S150 of FIG. S160 that follows
Let q = Bros (q), in this case Bros
Since (4) is 0 from FIG. 7, q = 0 and S120
Returning to step S180, “＼” is output. Therefore, only when “chi” and “＼” are output, S
Move to 190. Then, S200, S210, S2
20 and the process proceeds to S230, where the compression (Child
(4)) Perform processing. As can be seen from FIG. 7, Child
Since (4) = 5, in S230, the compression (5) process is performed as a recursive read process.

In the compression (5) processing, q = p = 5 in S110 of FIG. 11, and Num (5) = n = 5 in S130.
8, n = 8 + 1 = 9 is set. Then, in S140, since the content of Label (5) is "W" indicating the end of the word, the flow shifts to S170. Then, in S170, “W” is output and Child (q) is output. In this case, since Child (5) = 1, this 1 is also output.

Then, at S160, q = Bros
(Q), since Bros (5) is 0 from FIG. 7, the process returns to S120 as q = 0, and since q = 0,
The flow shifts to S180, where "@" is output. In the compression (5) process, “W”, “1”, and “＼”
Is output.

In S190 to which the process proceeds after the process in S180, q = p = 5, and the process proceeds to S200 and S210. In S210, since Label (5) is "W", the process directly proceeds to S240. Then, q = Bros (q) is set in S240, but since Bros (5) is 0 from FIG. 7, the process returns to S200 with q = 0, and S20
Since q = 0 at 0, the compression (5) processing ends.

Since the compression (5) processing is a recursive readout processing executed in S230 during the compression (4) processing described above, when the compression (5) processing is completed, the process proceeds to S240 during the compression (4) processing. I do. In this S240, q = Bro
s (q), since Bros (4) is 0 from FIG. 7, the process returns to S200 as q = 0, and q = S in S200.
Since it is 0, the compression (4) process ends.

Since the compression (4) process is a recursive readout process executed in S230 during the compression (3) process, when the compression (4) process is completed, the process proceeds to S240 during the compression (3) process. I do. In this S240, q = Bro
Although s (q) is set, since Bros (3) is 6 from FIG. 7, q = 6 and the process returns to S200. In S200, q =
6, the flow shifts to S210, and since Label (6) is "cho" and not "W", the flow shifts to S220.

In S220, Num (Child (6))
Is greater than 0, but Child
(6) is "7" from FIG. At this point Num
Since (Child (6)) = Num (7) is 0,
The flow shifts to S230, where the compression (Child (6)) process, that is, the compression (7) process is performed as a recursive read process.

In the compression (7) process, Num (7) = n = 9 and n = 9 + 1 = 10 are set in S130 of FIG. Then, "U" of Label (7) is output in S150. In subsequent S160, q = Bros (q). In this case, since Bros (7) is 0 from FIG.
The process returns to S120 with q = 0, and shifts to S180 to output “＼”. Therefore, the process proceeds to S190 only when “U” and “＼” are output.

Then, in S190, q = 7 is set, and in S20
The process proceeds to 0, S210, and in S220, Num
It is determined whether (Child (7)) is greater than 0. Child (7) is “5” from FIG.
Since m (5) is set to “8” in S130 of the compression (5) process, S220 in this case is affirmatively determined, and the process proceeds to S250.

Therefore, "J" is output in S250, and Num (Child (7)) = Num (5)
= 8, and then the flow shifts to S240. This is the part "J, 8" in the compressed dictionary data of FIG.
In S240, q = Bros (q) is set.
Since (7) is 0 in FIG. 7, q = 0 and S200
Returning to S200, since q = 0, the compression (7) process ends.

Since the compression (7) processing is a recursive readout processing executed in S230 during the compression (3) processing described above, when the compression (7) processing is completed, the process proceeds to S240 during the compression (3) processing. I do. At this time, considering that the compression (7) in S230 is compression (Child (6)), q in this case is 6. In S240, q = Bro
s (q), since Bros (6) is 0 from FIG. 7, the process returns to S200 with q = 0, and q =
Since it is 0, the main compression (3) processing ends.

The compression (3) process is a recursive read process executed in S230 during the above-described compression (2) process.
The process proceeds to S240 during processing. At this time, considering that the compression (3) processing in S230 is a compression (Child (2)) processing, q is set to Bros (2) in S240, but Bros (2) is 8 in FIG. Therefore, the process returns to S200 as q = 8, and since q = 8 in S200,
Move to S210. In the processing of S210, Label
Since (8) is "mi" and not "W", the flow shifts to S220.

At S220, Num (ChildN
It is determined whether (8)) is greater than 0. Chil
d (8) is “9” from FIG. 7, but Num (9) is 0.
Therefore, the process shifts to S230 and the compression (Child
(8)) Perform processing. Since Child (8) = 9, the compression (9) process is performed as a recursive read process in S230.

In the compression (9) process, q = p = 9 in S110 of FIG. 11, and Num (9) = n =
10, n = 10 + 1 = 11 is set. And S14
In the case of 0, since the content of Label (9) is “W” indicating the end of the word, the flow shifts to S170. And S1
At 70, “W” is output and Child (q) is output. In this case, since Child (9) = 2, this 2 is also output.

Then, at S160, q = Bros
(Q), since Bros (9) is 10 from FIG. 7, the process returns to S120 with q = 10. In S120, which returns after the processing of S160, since q = 10 and not 0, the flow shifts to S130 and Num (10)) = n =
11, n = 11 + 1 = 12. At S140, the content of Label (10) is changed to "W" indicating the end of the word.
However, since the content of Label (10) is “mi” and not “W”, the flow shifts to S150 and Label
(10) It outputs itself, that is, "".

In the subsequent S160, q = Bros (q) is set. Since Bros (10) is 0 from FIG.
It returns to S120 as 0. In S120, which returns after the processing of S160, q = 0, so that the flow shifts to S180, and “＼” is output.

In the compression (9) processing, “W”, “2”, “ki” and “＼” are output by the processing so far. S
In S190, which shifts after the process of 180, q = p = 9, and the shift shifts to S200 and S210. In S210, since Label (9) is “W”, the process goes to S
Move to 240. Then, in S240, q = Bros
(Q), since Bros (9) is 10 from FIG. 7, the process returns to S200 with q = 10. In subsequent S210, since Label (10) is "K" and not "W", the flow shifts to S220.

In S220, Num (Child (1
0)) is greater than 0, but Chil
d (10) is "11" from FIG. At this point, N
Since um (Child (10)) = Num (11) is 0, the flow shifts to S230 to compress (Child (1)
0)), that is, the compression (11) process is performed as a recursive read process.

In the compression (11) process, S110 of FIG.
And q = p = 11, and Num (11) at S130.
= N = 12, n = 12 + 1 = 13. And
In S140, since the content of Label (11) is "ta" and not "W", the flow shifts to S150 and Label
(11) Output itself, that is, “ta”.

In S160, q = Bros (q) is set. Since Bros (11) is 12 in FIG.
= 12 and the process returns to S120. Since q = 12 is not 0, the process proceeds to S130, where Num (12) = n =
13, n = 13 + 1 = 14 is set.

Then, in S140, Label (1
Since the content of 2) is “W” indicating the end of the word, S17
Move to 0. Then, in S170, “W” is output and Child (q) is output. In this case, Ch
Since ld (12) = 3, this 3 is also output.

In the subsequent S160, q = Bros
(Q), since Bros (12) is 0 from FIG. 7, the process returns to S120 with q = 0. In S120 returned after the processing of S160, since q = 0, S180
Move to. Then, “＼” is output in S180.
Therefore, only “ta” and “＼” are output and S19
Move to 0.

In S190, q = p = 11 is set again. Since q is not 0, the flow shifts to S210. In S210, La
Since Bel (11) is “ta” and not “W”, S220
Move to. In S220, Num (Child (1
It is determined whether 1)) is greater than 0. Child
(11) is “12” from FIG. 7, but at this time Nu
Since m (12) is still “0”, the flow shifts to S230, where the compression (Child (11)) process, ie, the compression (12) process is performed as a recursive read process.

In the compression (12) process, S130 of FIG.
Is set to Num (12) = n = 13, n = 13 + 1 = 14. Then, in S140, Label (12)
Is "W" indicating the end of the word, the process proceeds to S170. Then, in S170, “W” is output and Child (q) is output. In this case, Chil
Since d (12) = 3, this 3 is also output.

Then, at S160, q = Bros
(Q), since Bros (12) is 0 from FIG. 7, the process returns to S120 as q = 0, and the process proceeds to S180 because q = 0, and outputs “＼”. By the processing up to this point, “W”, “3”, and “＼” are output.

In S190, q = p = 12, and
The process proceeds to S200 and S210.
Since Bel (12) is “W”, S240 is performed as it is.
Move to. Then, in S240, q = Bros (q). Since Bros (12) is 0 from FIG.
= 0 and the process returns to S200. Since q = 0 in S200, the compression (12) process ends.

The compression (12) process is a recursive read process executed in S230 during the compression (11) process described above.
1) The process proceeds to S240 during processing. In this S240, q
= Bros (q), but since Bros (11) is 0 from FIG. 7, the process returns to S200 with q = 0, and S20
Since q = 0 at 0, the compression (11) process ends.

Since the compression (11) processing is a recursive readout processing executed in S230 during the compression (9) processing, the compression (9) processing is completed when the compression (11) processing is completed.
The process proceeds to S240 during processing. At this time, considering that the compression (11) in S230 is compression (Child (10)), q = Bros (10) in S240, but since Bros (10) is 0 from FIG. q = 0
Then, the process returns to S200, and since q = 0 in S200, the compression (9) process ends.

This compression (9) processing is a recursive readout processing executed in S230 during the above-described compression (2) processing. Therefore, upon completion of the compression (9) processing, the flow shifts to S240 during the compression (2) processing. I do. At this time, considering that the compression (9) processing in S230 is compression (Child (8)), in S240, q = Bros (8) is set.
Since ros (8) is 0 from FIG. 7, it is assumed that q = 0 and S
Returning to S200, since q = 0 in S200, the compression (2) process ends.

When the above-described compression (2) processing is completed, the numbers up to “ho, han, ma, chi, W, cho, u, mi, W, ki, ta, a” shown in FIG. Num (i) for each of the labels 1 to 13 is “1,3,5,7,8,6,9,4,1” as shown at the point B in FIG.
0, 11, 12, 13, 2 ". When the labels are rearranged in this order and “W” indicating the end of the word and the word number, and “J” indicating the merge and the label number of the merge destination are added, the compressed dictionary data at this time becomes As shown in the case of point B in FIG.
＼, Ma, Cho, ＼, Chi, ＼, W, 1, ＼, U, J,
8, W, 2, ＼, ＼, ＼, W, 3, ＼ ”.

Next, the progress up to the point C will be examined in order. A description will be given after the above-mentioned compression (2) processing is completed. The compression (2) process is a recursive read process executed in S230 during the compression (1) process described above. Therefore, when the compression (2) process ends, S240 during the compression (1) process is completed.
Move to. In this S240, q = Bros (q), but since Bros (1) is 13 from FIG.
The process returns to S200 as 13. In step S200, since q = 13, the flow shifts to step S210, and Label (13) returns
Is not "W", the flow shifts to S220.

In S220, Num (Child (1
3)) determines whether or not is greater than 0.
d (13) is “14” from FIG. At this point, N
Since um (Child (13)) = Num (14) is 0, the flow shifts to S230 to compress (Child (1)
3)) The process, ie, the compression (14) process, is performed as a recursive read process.

In the compression (14) process, S110 of FIG.
Is set to q = p = 14, and Num (1) is set in S130.
4) Set = n = 14, n = 14 + 1 = 15. In the following S140, since the content of Label (14) is “I” and not “W” indicating the end of the word, in S150,
"i" of abel (14) is output. In subsequent S160, q = Bros (q). In this case, Bros (1
Since 4) is 0 from FIG. 7, q = 0 is returned to S120, and since q = 0, the flow shifts to S180 to output “＼”.

Therefore, the process proceeds to S190 only when “I” and “$” are output. In S190, q = p =
The process proceeds to S210 because q is not 0. S2
At 10, since Label (14) is “I” and not “W”, the flow shifts to S220. In S220, Num (C
hold (14)) is greater than zero. Child (14) is “15” from FIG. 7,
At this point, Num (15) is still “0”, so S
The process proceeds to 230, and the compression (Child (14)) process, that is, the compression (15) process is performed as a recursive read process.

In the compression (15) process, S110 in FIG.
, Q = p = 15, and Num (15) = in S130.
Set n = 15, n = 15 + 1 = 16. And S
At 140, since the content of Label (15) is "O" and not "W" indicating the end of the word, La at S150
The “O” of the bell (15) is output. In subsequent S160, q = Bros (q). In this case, Bros (1
Since 5) is 0 from FIG. 7, the process returns to S120 as q = 0, and the process proceeds to S180 because q = 0, and “＼”
Is output.

Therefore, the process proceeds to S190 only when "O" and "$" are output. In S190, q = p =
The process proceeds to S210 because q is not 0. S2
At 10, since Label (15) is “O” and not “W”, the flow shifts to S220. In S220, Num (C
hold (15)) is greater than zero. Child (15) is “16” from FIG. 7,
At this point, Num (16) is still “0”, so S
The process proceeds to 230, and the compression (Child (15)) process, that is, the compression (16) process is performed as a recursive read process.

In the compression (16) process, S110 of FIG.
, Q = p = 16, and Num (16) = in S130.
Set n = 16, n = 15 + 1 = 17. And S
At 140, since the content of Label (16) is “I” and not “W” indicating the end of the word, La at S150
The "i" of the bell (16) is output. In subsequent S160, q = Bros (q). In this case, Bros (1
Since 6) is 0 from FIG. 7, q = 0 is returned to S120, and since q = 0, the flow shifts to S180, and “＼”
Is output. Therefore, the process proceeds to S190 only when “i” and “＼” are output.

In S190, q = p = 16 is set again. Since q is not 0, the flow shifts to S210. In S210, La
Since Bel (16) is “I” and not “W”, S220
Move to. In S220, Num (Child (1
6) Determine if is greater than 0. Child
(16) is “17” from FIG. 7, but at this time Nu
Since m (17) is still “0”, the flow shifts to S230, where the compression (Child (16)) process, ie, the compression (17) process is performed as a recursive read process.

In the compression (17) process, S110 of FIG.
And q = p = 17, and Num (16) = in S130.
Set n = 17, n = 15 + 1 = 18. And S
At 140, since the content of Label (17) is "W" indicating the end of the word, the flow shifts to S170 to output "W" and to output Child (17) = 4. In S160, q = Bros (q). In this case, B = Bros (q).
Since ros (17) is 0 from FIG. 7, the process returns to S120 with q = 0, and the process proceeds to S180 because q = 0, and outputs “＼”.

Therefore, "W", "4", and "$" are output, and the flow shifts to S190. In S190, q = p again
= 17, and since q is not 0, the flow shifts to S210. S
At 210, since Label (17) is “W”, S
Move to 240. In S240, q = Bros (q)
However, since Bros (17) is 0 from FIG.
The process returns to S200 with q = 0, and since q = 0 in S200, the compression (17) process ends.

The compression (17) process is a recursive read process executed in S230 during the compression (16) process described above.
6) The process proceeds to S240 during processing. In this S240, q
= Bros (q), but since Bros (16) is 0 from FIG. 7, q = 0 is returned to S200, and S20
Since q = 0 at 0, the compression (16) process ends.

The compression (16) process is a recursive readout process executed in S230 during the above-described compression (15) process.
5) The process proceeds to S240 during processing. In this S240, q
= Bros (q), but since Bros (15) is 0 from FIG. 7, the process returns to S200 with q = 0, and S20
Since q = 0 at 0, the compression (15) process ends.

Similarly, the compression (15) process is a recursive read process executed in S230 during the above-described compression (14) process.
4) The process proceeds to S240 during processing. In this S240, q
= Bros (q), but since Bros (14) is 0 from FIG. 7, the process returns to S200 with q = 0, and S20
Since q = 0 at 0, the compression (14) process ends.

Since the compression (14) processing is a recursive readout processing executed in S230 during the compression (1) processing described above, when the compression (14) processing is completed, S240 during the compression (1) processing is completed. Move to. At this time, S230
Considering that the compression (14) process in step (14) is compression (Child (13)), in S240, q = Bros (1
7), since Bros (13) is 0 from FIG. 7, the process returns to S200 as q = 0, and q = 0 in S200.
Therefore, the compression (1) process ends.

When the above-described compression (14) processing is completed, the compression (1) processing is all completed, and the “compression, (1) processing shown in FIG. , Ta,
Num (i) for each of the labels Nos. 1 to 17 of “A, I, O, I, W” is “1, 3, 5, 7, 8, 6, 6,” as shown at the point C in FIG. 9, 4, 10, 11,
12, 13, 2, 14, 15, 16, 17 ". Then, while sorting the labels in this order,
When "W" indicating the end of the word and the word number, and "J" indicating the merge and the label number of the merge destination are added, as shown in the case of the point C in FIG. "N, Mi, ＼, Ma, Cho, ＼, Chi, ＼, W, 1,
＼, ＼, J, 8, W, 2, ＼, ＼, ta, ＼, W, 3,
＼, お, ＼, ＼, ＼, ＼, W, 4, ＼ ”.

In the above, the specific processing at the time of converting the graph dictionary data into the compressed dictionary data has been described. As described above, after the conversion into the compressed dictionary data, Performs data compression and stores the compressed "compressed dictionary data" together with map data and other data in a map C for a navigation system.
Write to D. Then, the data in the map CD is read into the control circuit 10 via the map data input device 6 (see FIG. 1), and is transferred to the voice recognition device 30 as described above.

Then, the dialog control unit 3 of the voice recognition device 30
The dictionary control unit 32c in 2 develops the transferred compressed dictionary data into data, further develops the dictionary, and writes it into the dictionary unit 31b of the speech recognition unit 31. At the time of compression, general data compression such as, for example, variable-length encoding is performed on the compressed dictionary data. Therefore, the data expansion may be performed by a decompression method corresponding to the compression method, and a detailed description thereof will be omitted.

On the other hand, dictionary expansion is also an expansion method corresponding to the second conversion processing at the time of compression. However, if this dictionary expansion method is described in a descriptive manner, "the syllables and the end of words in the compressed dictionary data are determined. When the identified identification data is rearranged according to the second scanning order, and there is a sibling at the vertex, based on the presence of identification data for distinguishing the sibling relationship from the others, the compressed dictionary data In addition to the identification data indicating the syllable and the end of the word, the data for identifying the child is set for each identification data indicating the end of the syllable and the word and the data is developed into the second graph dictionary data.

In consideration of the conversion process in the second conversion process in the above-mentioned word dictionary data compression method, conversely, for each of the syllables in the compressed dictionary data and the identification data indicating the end of the word, one's own The data to identify the child is restored. That is, the syllables (each label corresponds in the above example) and the identification data indicating the end of the word (“＼” in the above example) in the compressed dictionary data are rearranged in the second scanning order. In addition, if there is a sibling at the vertex, identification data for distinguishing the sibling relationship from the others is set, and in consideration of this point, the tree structure similar to the tree structure shown in FIGS. The child can be easily analyzed from the parent-child relationship of the vertices of the directed graph.

Since it is easier to understand by referring to the specific example, a specific example of the second graph dictionary data is shown as a result of the text dictionary data shown in FIGS. 5 to 7 being converted by the second conversion processing. It is shown in FIG. As can be seen from FIG.
Each label (Labe, "ho, a, u, mi, ...")
For 1), data for identifying its own child (Child) is set by a label number. When there is a sibling, identification data for distinguishing the sibling relationship from others is set as “1” in the Last item.

As can be seen from FIG. 9, the second graph dictionary data is not necessarily the same as the state of the graph dictionary data (see FIG. 7) obtained by the first conversion processing by the above-described compression method. Absent. In the above-described graph dictionary data of FIG. 7, each label has data for identifying its own child and sibling. However, in the second graph dictionary data of FIG. It is not necessary to have data for identifying each sibling of each label.

In any case, in addition to the data of each label, it is possible to have, at a minimum, data for identifying one's own child and identification data for distinguishing a sibling from each other. The second dictionary data can be sufficiently used as a word dictionary.

Then, the compressed dictionary data shown in FIG.
A specific process for expanding the data into the second graph dictionary data shown in FIG. As shown in FIG. 16, when the process is started, a variable n is set to an initial value 1 in a first step S1010, and a developing process is executed in a succeeding S1020.

The details of this expansion processing will be described with reference to FIGS. In the first step S1110 in FIG. 17, a variable t = n is set. That is, initially, the initial value n = 1. Then, in the subsequent S1120, x = next input data is set, and in S1130, it is determined whether or not it is “J”.

If it is determined in S1130 that the input data x is “J”, the flow shifts to S1180, where x = next input data, and in S1190, this processing is performed using x as a return value. To return. On the other hand, input data x
If is not “J”, the flow shifts to S1140, and it is determined whether or not “＼” in S1140. If it is determined that the input data x is “$”, the flow shifts to S1210, sets Last (n−1) to 1, and then returns to FIG.
To S1220. The processing after S1220 will be described later.

If it is determined that the input data x is not "$", the flow shifts to S1150, where it is determined whether or not the input data x is "W". If it is determined that the input data x is “W”, the process proceeds to S1200, in which Label (n) is set to “W”, and Child (n) is set to the next input data. The process moves to S1170.

On the other hand, if it is determined that the input data x is not "W", the flow shifts to S1160, where Label is changed to S1160.
(N) = x, that is, the next input data set as x in S1120, and after setting Child (n) = 0, the process proceeds to S1170. In S1170, La
st (n) is set to 0, n is incremented (n = n + 1), and the process returns to S1120.

Next, the processing after S1220 in FIG. 18 will be described. In step S1220, the variable k is set to k = t.
At 0, Child (k) is compared with 0. Child
If (k) is not 0, the process directly proceeds to S1250, but if Child (k) = 0, the process of S1240 is executed, and then the process proceeds to S1250. Note that S1240
Is a process of executing the expansion process of FIGS. 17 and 18 as a recursive read process, and setting a return value when the process is returned to Child (k).

In S1250, Last (k) is compared with 1, and if Last ((k) is not 1, k is incremented (k = k + 1) in S1260 and the process returns to S1230, but if Last (k) = 1, If there is, the process returns in S1240 with t as a return value.

The above is the description of the flowchart showing the expansion processing. For easier understanding, the progress of developing the compressed dictionary data shown in FIG. 8 into the second graph dictionary data shown in FIG. 9 will be described. Will be described further.
FIG. 19 conceptually shows the contents of the expansion processing in S1020 of FIG. In other words, in the expansion processing, the expansion processing shown in S1240 of FIG. 18 is executed as a recursive reading processing while the processing is being executed.

A point A in FIG. 19 indicates a point in time before the expansion processing by recursive reading in S1240 is executed.
Similarly, point B indicates a point in time when the expansion processing by recursive reading in S1240 returns in S1270, and point C indicates a point in time when the expansion processing is completed to the end. FIG. 20 shows A
FIG. 21 shows a state at point B (intermediate progress), FIG.
2 indicates a state at point C.

FIG. 23 shows points A, B, and C described above.
Shows uninput data at a point. First, the progress up to the point A will be sequentially examined. In S1110 of FIG. 17, t = n
= 1, and the first input data “HO” is set as x in S1120. "J", "＼", "W"
Therefore, in S1160, Label (1) =
"H" and Child (1) = 0. Further, Last (1) is set to 0 in S1170.

Since n = 1 + 1 = 2 in step S1170, the process returns to step S1120, and the second input data “A” set in x is set to “Label” in step S1160.
1 (2) = “A”, and Child (2) = 0. Further, Last (2) is set to 0 in S1170.

At S1170, n = 2 + 1
Since it is set to 3, the process returns to S1120 and moves to S1210 for the third input data “＼” set to x in the determination of S1140. Then, since Last (n-1) = 1, that is, n = 3 in S1210,
Last (2) is set to 1.

As a result of the processing up to this point, as shown in FIG.
The ild is also “0”, and the label “a” has Last as “1” and Child as “0”. Next, the progress up to the point B will be examined in order.

After the processing in S1210 described above, the flow shifts to S1220 in FIG. In S1220, the variable k = t = 1
Then, in S1230, Child (1) is compared with 0. At this point, as shown in FIG.
Since (1) = 0, the flow shifts to S1240, and FIG.
7. The expansion processing of FIG. 18 is executed as a recursive read processing.

In the first step S1110 of FIG. 17 as a recursive read process, t = n = 3. this is,
This is because n = 3 has been incremented in the second S1170 executed during the above-described expansion processing. S1120
Then, the next input data is set as x. this is,
"N", which is the first data of the uninput data at point A in FIG. 23 and corresponds to the fourth data in the compressed graph dictionary data of FIG. 8, is set.

This “n” can be either “J” or “＼”,
Since it is not "W", Label (3) is determined in S1160.
= “HO” and Child (3) = 0. Further, Last (3) is set to 0 in S1170. Since n = 3 + 1 = 4 in S1170, S11
Returning to step 20, for the next input data "mi" to be set to x, in step S1160, Label (4) is set to "mi" and Child (4) is set to 0. Further, S1
At 170, Last (4) is set to 0. This S1170
Is set to n = 4 + 1 = 5.

After S1170, the process returns to S1120 to set the next input data “＼” to x. With respect to the input data “＼”, the process shifts to S1210 based on the judgment in S1140. Then, in S1210, Last (n-1)
= 1, that is, n = 5, so Last (4) is set to 1.

By the processing up to this point, Last is “0” and Child is “0” for the third label “n”, and Last is “1” and Child is “0” for the fourth label “Mi”. Becomes In the following S1220, the variable k is set to t = 3, and in S1230, Child is set.
Compare (3) with 0. At this point, Child
Since (3) = 0, the flow shifts to S1240, and FIG.
7. The expansion processing of FIG. 18 is executed as a recursive read processing.

In the first step S1110 of FIG. 17 as a recursive read process, t = n = 5. Similar to the processing for the three input data “n, mi, ＼” described above, the subsequent processing for the three input data “ma, cho, ＼” is performed by recursive readout, whereby the fifth label is obtained. For "ma", Last is "0" and Child is "0", and the sixth label "cho"
For, Last is “1” and Child is “0”.

In the following S1220, the variable k = t = 5, and
In S1230, Child (5) is compared with 0. At this time, since Child (5) = 0, S124
0, and executes the expansion processing of FIGS. 17 and 18 as a recursive read processing. FIG. 1 as a recursive read process
In the first step S1110 of No. 7, t = n = 7. Similarly to the above-described processing, in the expansion processing by recursive readout for two input data of “chi, ＼” following “ma, cho, ＼”, the seventh label “chi” is L
Ast is “1” and Child becomes “0”.

Then, in the same manner as described above, the variable k = t = 7 in S1220 to which the processing shifts after S1210, and Child (7) is compared with 0 in S1230. At this point, Ch
Since ld (7) = 0, the process proceeds to S1240,
The expansion processing of FIGS. 17 and 18 is executed as a recursive read processing.

In the first step S1110 of FIG. 17 as a recursive read process, t = n = 8. Similarly to the above-described processing, in the expansion processing by recursive reading for three input data “W, 1, ＼” following “chi, ＼”, in the first step S1110 in FIG.
8 is assumed. If the input data is “W”, S115
In order to shift to S1200 in the determination process of 0, S120
0, Label (8) = “W”, and C
The next input data “1” is set as “hold (8)”. Then, since the next input data is “$”, Last (8) = 1 in S1210. Therefore, for the eighth label "W", Last is "1" and Child is also "1".

In this case, the process proceeds to S1210 after S1210.
At 20, k = t = 8, and at S 1230, Chil
Since d (k) = Child (8) = 1 and not 0, the flow shifts to S1250. Then, since Last (8) = 1, the flow shifts to S1270, and returns with t (= 8) as a return value.

The expansion processing by recursive reading of the input data “W, 1, ＼” is performed by input data “C, ち”, where t = n = 7 in the first step S1110 in FIG.
Is a recursive readout process in S1240 during the expansion process for, and if t = 8 is returned as a return value, then Child (k) = 8 in S1240. That is, since the process of k = t = 7 is being performed in S1220, Child (7) = 8 as a result.

Then, in the following S1250, Last
Since (k) = Last (7) = 1, the flow shifts to S1270, and returns with t (= 7) as a return value. The expansion processing by recursive reading for the input data “chi, ＼” is performed in the first step S1110 in FIG.
5, since it is the recursive readout processing in S1240 during the expansion processing for the input data “ma, cho, ＼”, if t = 7 is returned as the return value,
At 1240, Child (k) = 7. That is, S
At 1220, k = t = 5, and as a result, Child
(5) = 7.

Then, in the following S1250, Last
Since (k) = Last (5) = 0, the flow shifts to S1260, where k = k + 1 = 6 is incremented and S123
Return to 0. In the following S1230, Child (k) = C
Since hold (6) = 0, the process proceeds to S1240,
The expansion processing of FIGS. 17 and 18 is executed as a recursive read processing.

FIG. 1 as an expansion process by recursive reading
In the first step S1110 of step 7, t = n = 9. In this case, x = next input data in S1120 is “U” following “W, 1” described above, and the next input data is “＼”.
Is “1” and Child becomes “0”. In this case, S1
In 170, n = 10 is incremented, and S1
In S1220 which is shifted after 210, k = t = 9, and
The processing after S1230 is performed.

Then, in S1230, Child (9)
= 0, so that the flow shifts to S1240. In S1240, expansion processing by recursive reading is performed on the two input data “J, 8” following the above “U, ＼”. In the first step S1110 in FIG. 17 as the expansion processing by recursive reading, t = n = 10. Then, the input data is “J”, and in the determination processing of S1130, S1
In step S1180, x = next input data = “8” in order to proceed to S180. Then, in S1190, the process returns with x = 8 as a return value.

Since the expansion processing by recursive reading for the input data “J, 8” is the recursive reading processing in S1240 during the expansion processing for the input data “U, ＼”, x = 8 is returned as a return value. In this case, Child (k) = 8 in S1240. The expansion processing for this “U, ＼” is performed in step S1220, where k = t
= 9, and as a result, Child
(9) = 8.

Then, in the following S1250, Last
Since (k) = Last (9) = 1, the flow shifts to S1270, and returns with t = 9 as a return value. The expansion processing by recursive readout for the input data “u, ＼” is performed during the expansion processing for the input data “ma, cho, ＼” with k = t = 5.
After returning to S1230, the process returns to S1230.
Since Child (6) = 0 at 30, the transferred S
This is the process at 1240. Therefore, in this case, the process is returned with t = 9 as the return value.
ild (6) = 9.

Then, in the subsequent S1250, Last
Since (k) = Last (6) = 1, the flow shifts to S1270, and returns with t (= 5) as a return value. The expansion processing by recursive reading for the input data “ma, cho, ＼” is performed in the first step S1110 in FIG.
= N = 3, and in S1220, S1220 during the expansion processing for the input data “n, mi, ＼” with k = t = 3.
Since the process is a recursive read process in step S40, if t = 7 is returned as a return value, Chi is returned in S1240.
ld (3) = 5.

Then, in the following S1250, Last
Since (k) = Last (3) = 0, the flow shifts to S1260, where k is incremented. That is, k = 3 + 1
= 4. Then, S123 is shifted to after S1260.
At 0, Child (3) is compared with 0. At this point, since Child (3) = 0, the flow shifts to S1240 to execute the expansion processing of FIGS. 17 and 18 as a recursive read processing.

FIG. 1 as an expansion process by recursive reading
In the first step S1110 of step 7, t = n = 10. In this case, x = next input data in S1120 is “W” following “J, 8” described above. That is, S12
At 00, Label (10) = “W” and further C
"2" which is the next input data is set as hold (10). In S1170, Last (10) =
0 and incremented to n = 10 + 1 = 11.

Then, in S1120, since the subsequent input data is "", the label data is sent in S1160.
(11) = “G”, Child (11) = 0, and in S1170, Last (11) = 0, and n = 11 + 1 = 12 is incremented.

Since the subsequent input data is “＼”,
Last (11) = 1 in S1210, and S122
Move to 0. Therefore, for the tenth label "W", Last is "0" and Child is "2", and for the eleventh label "K", Last is "1" and Child is "0".

At S1220, k = t = 10, and S
Processing after 1230 is performed. Then, in S1230, since Child (10) = 2, the flow shifts to S1250. In S1250, Last (k) = Last (1
0) = 0, the flow shifts to S1260, and k = k + 1
= 11 and returns to S1230. The following S
In 1230, Child (k) = Child (11)
= 0, the flow shifts to S1240, and FIGS.
Is executed as a recursive read process.

FIG. 1 as an expansion process by recursive reading
In the first step S1110 of step 7, t = n = 12. In this case, x = next input data in S1120 is “ta” following “ki, ＼” described above. That is, S11
At 60, it is assumed that Label (12) = “ta” and further C
set hill (12) = 0. Also, S1170
Then, Last (12) = 0, and n = 12 + 1 = 13 is incremented.

Then, in S1120, since the subsequent input data is "$", the last data is input in S1210.
(12) = 1, and the flow shifts to S1220. Therefore, for the twelfth label “ta”, Last is “1” and Child is “0”.

At S1220, k = t = 12, and S
Processing after 1230 is performed. Then, in S1230, since Child (12) = 0, the flow shifts to S1240, and the expansion processing in FIGS. 17 and 18 is executed as a recursive read processing. FIG. 1 as expansion processing by recursive reading
In the first step S1110 of step 7, t = n = 13. In this case, x = next input data in S1120 is “W” following the above “ta, ＼”. That is, S12
At 00, Label (13) = “W”, and further C
hold (13) = next input data = 3 is set.
In S1170, Last (13) = 0, and n =
13 + 1 = 14 is incremented.

Then, in S1120, since the subsequent input data is “＼”, the last data is input in S1210.
(13) = 1, and the flow shifts to S1220. Therefore, for the thirteenth label “W”, Last is “1” and Child is “3”.

In S1220, k = t = 13, and S
Processing after 1230 is performed. Then, in S1230, since Child (13) = 3, the flow shifts to S1250. In S1250, Last (k) = Last (1
3) Since = 1, the flow shifts to S1270, and returns with t = 13 as a return value.

The expansion processing by recursive reading of the input data “W, 3,...” Is t = n = 12 in the first step S1110 in FIG. 17, and k = t in S1220.
= 12, since it is a recursive readout process in S1240 during the expansion process for the input data “ta, ＼”
If t = 13 is returned as a return value, S1
At 240, Child (12) = 13.

Then, in the following S1250, Last
Since (k) = Last (12) = 1, S1270
And returns with t = 12 as the return value. The expansion processing by recursive reading of the input data “ta, ＼” is performed by setting k = t = 10 in step S1220 of FIG. 18, incrementing k = k + 1 = 11 in step S1260, and returning to step S1230.
Since (11) = 0, the recursive readout process has been performed in S1240 after the transition, and if t = 12 is returned as a return value, then Child (1) is returned in S1240.
1) = 12.

Then, in the following S1250, Last
Since (k) = Last (11) = 1, S1270
Move to. In this case, since k = t = 10 in S1220, the process returns with t = 10 as a return value. In this process, t = n = 3 in the first step S1110 in FIG. 17, and k = t in S1220.
During the expansion process for the input data “n, mi, ＼” for which = 3, k = 3 + 1 = 4 is further incremented in S1260, and the process returns to S1230.
Since it is the recursive readout process in S1240 executed because ld (4) = 0, if the process is returned with t = 10 as the return value, the process proceeds to S1240.
(4) = 10.

Then, in S1250, Last
Since (k) = Last (4) = 1, the flow shifts to S1270. In this case, since k = t = 3 in S1220, the process returns with t = 3 as a return value. The expansion processing by recursive reading here is performed by setting t = n = 1 in the first step S1110 in FIG.
In the case of 0, since the expansion processing is being performed on the input data “ho, a, ＼” with k = t = 1,
If the process returns with t = 3 as the return value, S12
At 40, Child (1) = 3.

As shown in FIG. 21, through the series of processing up to this point, for each of “ho” of label number 1 to “W” of label number 13, the value of Last is either “0” or “1”. Is set, and for Child,
A value other than “0” is set for the label number 2 other than “A”. In other words, the number of the label of the relationship that becomes a child to oneself is given.

Next, the progress up to the point C will be examined in order. After the processing in S1240 described above, the flow shifts to S1250. In S1250, Last (k) = Last (1)
Since = 0, the flow shifts to S1260, where k is incremented. That is, k = 1 + 1 = 2. And S
In S1230 which is shifted to after 1260, Child (k)
= Child (2) = 0, the flow shifts to S1240, and the expansion processing in FIGS. 17 and 18 is executed as a recursive read processing.

In the first step S1110 in FIG. 17 as the expansion processing by recursive reading in this case, t = n
= 14. In this case, x = next input data in S1120 is “I” following “W, 2, ＼” described above.
That is, in S1160, Label (12) = “i”
And set Child (14) = 0. In S1170, Last (14) = 0 and n = 1
4 + 1 = 15 is incremented.

Then, in S1120, since the subsequent input data is “$”, the last data is inputted in S1210.
(14) = 1, and the flow shifts to S1220. Therefore, for the fourteenth label “I”, Last is “1” and Child is “0”.

In S1220, k = t = 14 and S
Processing after 1230 is performed. In S1230, since Child (14) = 0, the flow shifts to S1240, and the expansion processing in FIGS. 17 and 18 is executed as a recursive read processing. In the first step S1110 in FIG. 17 as the expansion processing by recursive reading in this case, t = n
= 15. In this case, x = next input data in S1120 is “O” following “I, ＼” described above. That is, in S1160, Label (15) = “O”, and Child (15) = 0 is set. In S1170, Last (15) = 0 and n = 1
Increment to 5 + 1 = 16. And S112
At 0, since the subsequent input data is “デ ー タ”, S
Last (15) = 1 at 1210, and S1220
Move to.

Therefore, for the fifteenth label “O”, Last is “1” and Child is “0”.
In S1220, k = t = 15, and the processes in S1230 and thereafter are performed. Then, in S1230, Child
Since (15) = 0, the flow shifts to S1240, and FIG.
7. The expansion processing of FIG. 18 is executed as a recursive read processing.

In this case, in the first step S1110 in FIG. 17 as the expansion processing by recursive reading, t = n
= 16. In this case, x = next input data in S1120 is “I” following the above “O, ＼”. That is, in S1160, Label (16) is set to "I", and Child (16) is set to 0. In S1170, Last (16) = 0 and n = 1
6 + 1 = 17 is incremented. And S112
At 0, since the subsequent input data is “デ ー タ”, S
Last (16) = 1 at 1210, and S1220
Move to.

Therefore, for the 16th label “I”, Last is “1” and Child is “0”.
In S1220, the process after S1230 is performed with k = t = 16. Then, in S1230, Child
Since (16) = 0, the flow shifts to S1240, and FIG.
7. The expansion processing of FIG. 18 is executed as a recursive read processing.

In the first step S1110 in FIG. 17 as the expansion processing by recursive reading in this case, t = n
= 17. In this case, x = next input data in S1120 is “W” following “I, ＼” described above. That is, the process proceeds from the determination processing of S1150 to S1200, and
At 1200, Label (17) = “W”, and Child (17) = next input data = 4 is set. In S1170, Last (17) = 0, and
Increment n = 17 + 1 = 18.

Then, in S1120, since the subsequent input data is “＼”, the last data is inputted in S1210.
(17) = 1, and the flow shifts to S1220. Therefore, for the 17th label "W", Last is "1" and Child is "4".

In S1220, k = t = 17 and S
Processing after 1230 is performed. Then, in S1230, since Child (17) = 4, the flow shifts to S1250. In S1250, Last (k) = Last (1
7) = 1, so the flow shifts to S1270, and returns with t = 17 as a return value.

The expansion processing by recursive reading of the input data “W, 4,...” Is t = n = 16 in the first step S1110 in FIG. 17, and k = t in S1220.
= 16, which is a recursive readout process in S1240 during the expansion process for the input data “i, ＼”
If t = 16 is returned as a return value, S1
At 240, Child (16) = 17.

Then, in the following S1250, Last
Since (k) = Last (16) = 1, S1270
And returns with t = 16 as the return value. The expansion processing by recursive reading for the input data “I, ＼” is performed in the first step S1110 in FIG.
15 and the recursive readout process in S1240 during the expansion process for the input data “O, と し た” in which k = t = 15 in S1220, so if t = 16 is returned as a return value, S1240 is returned. Child at
(15) = 16.

Then, in the following S1250, Last
Since (k) = Last (15) = 1, S1270
And returns with t = 15 as a return value. The expansion processing by recursive reading for the input data “O, ＼” is performed in the first step S1110 in FIG.
14 and the recursive readout process in S1240 during the expansion process for the input data “I, と し た” in which k = t = 14 in S1220, so if t = 15 is returned as a return value, S1240 is returned. Child at
(14) = 15.

In this processing, t = n = 1 in the first step S1110 in FIG. 17, and k = k in S1220.
During the expansion process for the input data “n, mi, ＼” with t = 1, k = 1 + 1 = 2 is further incremented in S1260, and the process returns to S1230.
Since the recursive readout process in S1240 was executed because hill (2) = 0, if t = 14 is returned as a return value, the process proceeds to S1240.
(2) = 14.

Then, in the following S1250, Last
Since (k) = Last (2) = 1, the flow shifts to S1270 to return to the present process. As shown in FIG. 22, by the series of processes up to this point, for each of “ho” of label number 1 to “W” of label number 17,
The value of Last is set to either “0” or “1”. In addition, for Child, in FIG. 21 showing the state at point B described above, “A” of label number 2 is “0”, but the value “14” is also set there. Of course, “Child” for each of “i” of the newly created label number 14 to “W” of the label number 17 is also set.

The program for executing the above-described word dictionary data compression method and expansion method is, for example, a workstation used for compression or a dialogue control unit 32 (see FIG. 1 and 2)
It is provided as a program started on the computer system side. In the case of such a program, for example, a floppy disk, a magneto-optical disk, a CD-ROM,
It can be used by storing it in a machine-readable recording medium such as a hard disk, loading it into a computer system as needed, and starting it. In addition, ROM
Alternatively, the program may be stored as a machine-readable recording medium such as a ROM or a backup RAM, and the ROM or the backup RAM may be incorporated in a computer system and used.

In the case of the car navigation system 2, instruction data related to navigation processing recognized by the voice recognition device 30 is sent to the control circuit 10, and navigation processing based on the instruction data is performed. Considering the nature of navigation, place name data forms the center of this instruction data. In addition to the unique property that there is a high possibility that there are a plurality of words having the same syllable data as described above, this place name data is newly added or updated due to the property that it is a part of the map data on the navigation device side. Many are assumed. That is, update data can be input through the map data input device 6.

In order to cope with the new addition or update of the map data, it is necessary to supply word dictionary data corresponding to the place name data to the speech recognition device 30 from outside. It is conceivable that new addition or updating of place names is basically realized as text format data.However, converting them into the above-described graph-format word dictionary data requires a large amount of memory and a high-performance CPU. Is done. Therefore, it is not practical to provide them on the side of the speech recognition device 30, and it is necessary to supply word dictionary data in the form of a graph that is created in response to the addition or update of them. However, since the data size is considerably large even if the dictionary data is in the form of a graph, considering the transfer of the word dictionary data from the control circuit 10 to the speech recognition device 30, the large-scale word dictionary data is transferred as it is. Doing so increases the communication time and is also not practical. For this reason, the communication time can be reduced by transferring the compressed dictionary data to the voice recognition device 30 after making the state of the compressed dictionary data.

As described above, the present invention is not limited to the above embodiments, and can be implemented in various forms without departing from the gist of the present invention. For example, the above-described compression processing shown in FIGS. 10 to 12 and the expansion processing shown in FIGS. 16 to 18 are examples of a case where the word dictionary data compression method and the expansion method of the present invention are realized. It is not limited to the processing itself. In other words, the compression processing shown in FIGS. 10 to 12 is performed in such a manner that “the syllable data and the word end Is assigned to each vertex according to the first scanning order, and is converted into graph dictionary data in which each vertex has data for identifying its own child and sibling. " is there. Similarly, the expansion processing shown in FIG. 16 to FIG. 18 is performed when “the syllables in the compressed dictionary data and the identification data indicating the end of the word are rearranged according to the second scanning order, and the vertex has a sibling. Based on the presence of identification data for distinguishing the sibling relationship from the others, in addition to the identification data indicating the syllable and word end in the compressed dictionary data, the identification This is an example in which the process for setting data for identifying the child of the child and expanding the data to the second graph dictionary data ”is specifically illustrated. In the above embodiment, the control circuit 10 of the navigation device writes word dictionary data to be transferred to the voice recognition device 30 to the navigation CD in the form of compressed dictionary data (see FIG. 3).
The conversion into the compressed dictionary data is executed, for example, at a workstation. This is based on the fact that a large amount of memory and a considerably high-performance CPU are required to make the word dictionary data in the graph format, so that it is practical to execute it on a workstation or the like. However, the control circuit 10 of the navigation device itself may execute the conversion processing to the compressed dictionary data.

Further, the compression processing of the present invention is the same as the above-described “...
.., Syllable data and identification data indicating the end of a word are assigned to each vertex according to the first scanning order, and as shown in “...”, The first scanning order is followed. A pre-order traversal was adopted as the scanning order of 1. Of course, other scans such as "inorder traversal" and "postorder traversal"
versal ", but these scans can be realized.

Although the above embodiment has been described as an example in which the present invention is applied to a car navigation system, there may be various other possible systems for inputting a user's instruction using a voice recognition function. The system can be similarly realized. However, in the case of a car navigation system, there is a limitation that the method is used for in-vehicle equipment, and it is necessary to input a destination or the like by a place name. The benefits are considered to be very large.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a car navigation system as an embodiment of the present invention.

FIG. 2 is a block diagram illustrating a configuration of a speech recognition unit and a dialogue control unit in the speech recognition device of the embodiment.

FIG. 3 is an explanatory diagram showing a procedure when generating compression dictionary data.

FIG. 4 is an explanatory diagram showing a specific example of text dictionary data.

FIG. 5 is an explanatory diagram showing a specific example of graph dictionary data.

FIG. 6 is an explanatory diagram showing a specific example of graph dictionary data by pointer expression.

FIG. 7 is an explanatory diagram showing a specific example of graph dictionary data in an array expression.

FIG. 8 is an explanatory diagram showing a specific example of compression dictionary data.

FIG. 9 is an explanatory diagram showing a specific example of second graph dictionary data.

FIG. 10 is a flowchart illustrating a compression process.

FIG. 11 is a flowchart showing a compression process.

FIG. 12 is a flowchart illustrating a compression process.

FIG. 13 is an explanatory diagram conceptually showing the contents of the compression (1) processing in S30 of FIG. 10;

FIG. 14 is an explanatory diagram showing an intermediate process when graph dictionary data is converted into compressed dictionary data.

FIG. 15 is an explanatory diagram showing the progress of the process when graph dictionary data is converted to compressed dictionary data.

FIG. 16 is a flowchart showing a developing process.

FIG. 17 is a flowchart illustrating an expansion process.

FIG. 18 is a flowchart showing a developing process.

FIG. 19 is an explanatory diagram conceptually showing the contents of a development process.

FIG. 20 is a diagram showing the progress in the process of expanding the compressed dictionary data into the second graph dictionary data.
It is explanatory drawing which shows the state in a point.

FIG. 21 is a diagram showing the progress of the expansion of the compressed dictionary data into the second graph dictionary data,
It is explanatory drawing which shows the state in a point.

FIG. 22 is a diagram showing the progress in the process of expanding the compressed dictionary data into the second graph dictionary data.
It is explanatory drawing which shows the state in a point.

FIG. 23 is an explanatory diagram showing non-input data at points A, B, and C in FIG. 19 when compressed dictionary data is expanded into second graph dictionary data.

[Explanation of symbols]

2 ... Car navigation system 4 ... Position detector 6 ... Map data input device 8 ... Operation switch group 10 ... Control circuit 12 ... External memory 14 ... Display device 15 ... Remote control sensor 15a ... Remote control 16 ... Geomagnetic sensor 18 ... Gyroscope 20 ... Distance sensor 22 GPS receiver 30 Voice recognition device 31 Voice recognition unit 31a Matching unit 31b Dictionary unit 32 Dialogue control unit 32a Post-processing unit 32b Communication control unit 32c Dictionary control unit 33 Voice synthesis unit 34: voice input unit 35: microphone 36: PTT switch 37: speaker

Claims (13)

[Claims] 1. A method for compressing word dictionary data in which a word group in which a plurality of words having the same syllable data constituting a word exist in a text format is compressed, wherein the text format word dictionary data is converted into a tree structure. The syllable data and the identification data indicating the end of a word are assigned to each vertex according to the first scanning order so that the number of paths reaching the vertices is not necessarily one, and the syllable data corresponds to a directed graph format similar to a tree structure. A first conversion process for converting each vertex into graph dictionary data having data for identifying its own child and sibling; and a tree structure for the graph dictionary data obtained by the first conversion process. The identification data indicating the syllable and the word end assigned to the vertices of the similar directed graph are stored in the first
If there is a sibling at the vertex while basically following the scanning order, the siblings are rearranged according to a second scanning order, and identification data for discriminating the self-sibling relationship from the others. And a second conversion process for converting the word dictionary data into compressed dictionary data having the following. 2. The method according to claim 1, wherein the first scanning order is a forward scanning. 3. The method according to claim 2, wherein data indicating that a predetermined syllable is the youngest is used as identification data for distinguishing a self-sibling relationship from another used in the second conversion process. 2. The method for compressing word dictionary data according to 1 or 2. 4. The method according to claim 1, wherein data indicating the number of siblings is used as identification data for distinguishing one's own sibling relationship from others used in the second conversion process. Word dictionary data compression method. 5. The word dictionary data converted into the compressed dictionary data by the second conversion processing is further compressed using a predetermined data compression method. How to compress the described word dictionary data. 6. A machine-readable recording medium, wherein the method for compressing word dictionary data according to claim 1 is recorded as a program to be executed by a computer system. 7. A method for expanding the compressed dictionary data compressed by the word dictionary data compression method according to claim 1, wherein the syllable and the word end in the compressed dictionary data. Is sorted in accordance with the second scanning order, and if there is a sibling at the vertex, the compression is performed based on the presence of identification data for distinguishing the sibling relationship from the others. In addition to the identification data indicating the syllable and the end of the word in the dictionary data, a word dictionary that sets data for identifying one's own child for each identification data indicating the syllable and the end of the word and develops the second graph dictionary data How to expand the data. 8. A method for expanding the compressed dictionary data compressed by the method for compressing word dictionary data according to claim 5, wherein the word dictionary data in a state compressed by the predetermined compression method is extracted. Expanding the compressed dictionary data in a state before being compressed by the predetermined compression method, wherein the identification data indicating the syllable and the end of the word in the expanded compressed dictionary data are rearranged according to the second scanning order. In addition, if there is a sibling at the vertex, based on the existence of identification data for distinguishing the sibling relationship from others, in addition to the identification data indicating the syllable and word end in the compressed dictionary data, The data for identifying the child is set for each of the identification data indicating the syllable and the end of the word, and the word dictionary data to be expanded into the second graph dictionary data is set. Open method. 9. A machine-readable recording medium, wherein the word dictionary data expansion method according to claim 7 or 8 is recorded as a program executed by a computer system. 10. A word dictionary storage means in which acoustic features of a plurality of words to be recognized are stored in advance for each word, and an acoustic analysis for extracting an acoustic feature by analyzing an externally input speech. Means for dividing the time-series data of the acoustic features extracted by the acoustic analysis means into data strings most similar to the acoustic features stored in the word dictionary storage means; And a voice recognition means for recognizing a word sequence of the input voice, and an output means for outputting a recognition result by the voice recognition means to an external device. An apparatus for inputting compressed dictionary data from an external device in a state where the word dictionary data is converted into the compressed dictionary data by the word dictionary data compression method according to any one of claims 1 to 5. A compressed dictionary data expanding unit that expands the compressed dictionary data input through the compressed dictionary data input unit into the second graph dictionary data by the word dictionary data expanding method according to claim 7 or 8. A speech recognition apparatus, characterized in that the word dictionary storage means stores word dictionary data expanded into the second graph dictionary data by the compression dictionary data expansion means. 11. A word dictionary storage unit in which acoustic features of a plurality of words to be recognized are stored in advance for each word, and an acoustic analysis for extracting an acoustic feature by analyzing an externally input speech. Means for dividing the time-series data of the acoustic features extracted by the acoustic analysis means into data strings most similar to the acoustic features stored in the word dictionary storage means; And a voice recognition means for recognizing a word sequence of the input voice, and an output means for outputting a recognition result by the voice recognition means to an external device. An apparatus for inputting compressed dictionary data from an external device in a state where the word dictionary data is converted into the compressed dictionary data by the word dictionary data compression method according to any one of claims 1 to 5. The word dictionary storage means stores the compressed dictionary data input from an external device by the compressed dictionary data input means, and further stores the compressed dictionary data when power is turned on or when speech recognition processing is executed. Speech recognition comprising compressed dictionary data expanding means for expanding the compressed dictionary data read from the means into the second graph dictionary data by the word dictionary data expanding method according to claim 7 or 8. apparatus. 12. A voice recognition device according to claim 10 and a navigation device, wherein said voice input means of said voice recognition device needs to be specified when said navigation device performs a navigation process. Is used for the user to input an instruction of certain predetermined navigation processing related data by voice, and the output unit is configured to output a recognition result by the voice recognition unit to the navigation device. Wherein the navigation device converts the word dictionary data into the compressed dictionary data by the word dictionary data compression method according to any one of claims 1 to 4. Compressed dictionary data storage means for storing in a state, and said compressed dictionary data storage means when a predetermined dictionary transfer is required Navigation system, characterized in that al the read compressed dictionary data, and a dictionary data transfer means for transferring the voice recognition device. 13. The navigation system according to claim 12, wherein the word dictionary data is converted into the compressed dictionary data by the word dictionary data compression method according to any one of claims 1 to 5. It is provided with a navigation-side compression dictionary data input unit input from an external device, and the compression dictionary data input via the navigation-side compression dictionary data input unit is stored in the compression dictionary data storage unit. A navigation system characterized by the following.