JP2008046260A - Voice recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition device which can register only the vocabularies actually used by a user as the target vocabularies for voice recognition. <P>SOLUTION: The signal processing unit 1 has an external memory 15 storing legitimate names as the target vocabularies for voice recognition. It creates other synonymous words from the legitimate names through the signal processor 11, and registers them in the external memory 15 as the vocabularies for voice recognition after evaluating how often they are used. It registers only the synonymous words estimated to be actually used often as the target vocabularies for voice recognition. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus that recognizes speech uttered by a user.

  Conventionally, as a speech recognition apparatus, techniques described in Patent Document 1 and Patent Document 2 below are known.

  Patent Document 1 describes a technique for creating a dictionary that is efficient and suitable for a user utterance style by defining a dictionary creation rule. Further, in Patent Document 1, the official name of the facility is input, the official name of the input facility is divided into morphemes, and the region name and a part of the official name of the facility are extracted and extracted from the divided morphemes. Registering a word obtained by concatenating a region name and a part of the extracted facility name with a concatenated word as a dictionary word.

Patent Document 2 describes that a long word is easily and surely recognized, and that the load on the user's utterance is reduced and the voice recognition adapted to the user's utterance style is performed. Also, in Patent Document 2, a long facility name is segmented to rephrase the recognition word, and the paraphrase word is registered in the dictionary so that the user can recognize the voice that shortens the stagnation of the user or the spoken word. Is described.
JP 2005-202198 A JP 2001/083982 A

  However, in the speech recognition technology described above, the number of recognized vocabularies increases more than necessary because the paraphrased word is added to the recognition dictionary even though the user does not know whether to actually use the paraphrase. As a result, there is a problem that the recognition rate is lowered.

  Therefore, the present invention has been proposed in view of the above-described circumstances, and an object thereof is to provide a speech recognition apparatus that can register only a vocabulary actually used by a user as a speech recognition target vocabulary. To do.

  The present invention is a speech recognition apparatus including speech recognition means for recognizing speech uttered by a user, the storage means storing a formal name as a speech recognition target vocabulary, and the formal name stored in the storage means A paraphrase word generating means for generating a paraphrase word, an actual usage evaluation means for evaluating the actual usage of the paraphrase word generated by the paraphrase word generating means, and the paraphrase word generated by the paraphrase word generating means as a speech recognition target vocabulary Registration means for registering in the storage means. In order to solve the above-described problem, such a speech recognition apparatus registers only paraphrased words that have been evaluated as having high actual usage by the practicality evaluation unit by the registration unit as a speech recognition target vocabulary.

  According to the speech recognition apparatus according to the present invention, since the paraphrase word having high actual usage is registered as the speech recognition target vocabulary, only the vocabulary actually used by the user can be registered as the speech recognition target vocabulary. There is no problem that the vocabulary for speech recognition increases more than necessary.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[First Embodiment]
The present invention is applied to, for example, the speech recognition apparatus according to the first embodiment configured as shown in FIG. This speech recognition apparatus is configured by connecting a signal processing unit 1 to a microphone 2, a touch panel display 3, a speaker 4, and an input device 5. This speech recognition apparatus registers only paraphrases that have a high degree of actual use (actual usage) by users when registering paraphrases for official names.

  In the signal processing unit 1, an A / D converter 12, a D / A converter 13, an amplifier 14, and an external storage device (storage means) 15 are connected to a signal processing device 11.

  The signal processing device 11 is supplied with the user's voice signal detected by the microphone 2 via the A / D converter 12. Further, the signal processing device 11 displays an operation name, a voice recognition result, and the like on the touch panel display 3 and inputs a user operation input signal from the touch panel display 3. Further, the signal processing device 11 supplies a voice signal to the speaker 4 via the D / A converter 13 and the amplifier 14 to give voice guidance of various information, and gives a notification voice for instructing to select an operation name. The announcement sound of the voice recognition result is emitted from the speaker 4.

  The signal processing device 11 includes a CPU (Central Processing Unit) 21 and a memory 22. The signal processing apparatus 11 uses the memory 22 as a work area, and the CPU 21 evaluates speech recognition processing (speech recognition means), processing to generate paraphrase words (paraphrase word generation means), and actual usage of paraphrase words. Processing (actual usage evaluation means), processing to register paraphrases (registration means) is performed.

  The input device 5 is operated when correcting the speech recognition result when a speech recognition result different from the speech intended by the user is generated by the speech switch 5a and the signal processing unit 1 which are operated when starting the speech recognition. And a correction switch 5b (correction instruction means). When the speech switch 5a and the correction switch 5b are operated, the operation is detected by the signal processing device 11. In addition, when the correction switch 5b is kept pressed for a certain period, the signal processing unit 1 terminates the processing by the voice emitted from the user halfway.

  The external storage device 15 is a speech recognition target vocabulary database in which formal name information of facilities, etc., speech recognition target vocabulary as paraphrase information and position information of the facilities are registered, and an operation name database in which operation names at the time of operation are registered. Is remembered. For example, as shown in FIG. 2, an operation name database 31 in which destination names, which are operation names for the signal processing unit 1, and operation names of higher layers such as search conditions are registered, addresses of lower layers of destinations, lower layers of facilities, etc. It consists of an operation name database 32 in which operation names are registered, and a speech recognition target vocabulary database 33 in which formal names and paraphrased words corresponding to lower layers of the operation name databases 31 and 32 are registered.

  For example, when the destination of the user is a facility name “Hokkaido University”, the operation name database 31 stores the operation name of the highest layer destination, search conditions, etc., and the operation name database 32 stores the lower layer of the destination. Assuming that operation names such as addresses and facilities are registered, the operation name “destination” is selected from the operation name database 31, and the operation name “facility name” is selected from the operation name database 32. The speech recognition target vocabulary database 33 is registered with the facility name “Hokkaido University” in the lowest layer for the operation name databases 31 and 32.

  Further, the external storage device 15 stores a speech recognition target vocabulary database in which a part of the formal name, paraphrase word, and operation name database is registered as a speech recognition target. This speech recognition target vocabulary database is rewritten by the signal processing device 11.

  Next, in the speech recognition apparatus described above, when the user wants to go to the facility with the official name “Hokkaido University” or display the map, the paraphrase “Hokkaido” is not registered in the speech recognition target vocabulary database. The operation at the time will be described.

  In such a voice recognition device, the voice input is accepted when the speech switch 5a is operated and the voice menu image 41 for voice input shown in FIG. The voice menu image 41 includes a voice recognition result display column 45 for displaying a voice recognition result by the signal processing unit 1 and a menu list 46 indicating operation names. The menu list 46 lists voice recognition target words that are operation names such as destinations and search conditions registered in the voice recognition target word database of the external storage device 15.

  In the state where the voice menu image 41 is displayed, the signal processing unit 1 allows the user to speak the operation name “destination” as shown in (1) of FIG. When the result is obtained, the signal processing unit 1 emits an announcement voice “Please give destination command” from the speaker 4 as shown in (2) of FIG. Is displayed on the touch panel display 3.

  In the state where the destination selection menu image 42 of FIG. 3B is displayed, the user selects the operation name “facility” included in the menu list 46 of the destination selection menu image 42 as shown in FIG. When a speech recognition result of “facility” is obtained by utterance and signal processing unit 1, an announcement voice “Please name the facility” is emitted from speaker 4 as shown in FIG. The facility name input image 43 shown in FIG.

  In the state where the facility name input image 43 of FIG. 3C is displayed, when the user speaks “Hokkaido University” as shown in FIG. 4 (5), the signal processing unit 1 is stored in the external storage device 15. Since “Hokkaido University” is not registered as a speech recognition target vocabulary in the speech recognition target vocabulary database, select the Kokubun station registered in the speech recognition target vocabulary database closest to the speech recognition result for “Hokkaido University”. As shown in FIG. 3D, the speech recognition result image 44 including the speech recognition result “Kokubun Station” is displayed in the speech recognition result display field 45. In this voice recognition result image 44, a “go to there” button 47 for generating a command for searching for an optimum route to go to the position indicated by the voice recognition result included in the voice recognition result display field 45 and the voice recognition result indicate. And a “view map” button 48 that generates a command to display a map near the location.

  When the user operates the correction switch 5b of the input device 5 as shown in FIG. 4 (7) in the state where the voice recognition result image 44 is displayed, the "4" is displayed from the speaker 4 as shown in FIG. “Speak again” is sounded and the facility name input image 43 of FIG. 3C is displayed again. Then, the steps (9) to (12) and (13) to (16) are performed as in (5) to (8) in FIG. 4, and in (17) in FIG. When speaking, since the official name “Hokkaido University” is registered as the speech recognition target vocabulary in the speech recognition target vocabulary database, the speech recognition result of “Hokkaido University” by the signal processing unit 1 is shown in FIG. It can be displayed in the speech recognition result display field 45 in the recognition result image 44.

  As described above, when “Hokkaido University”, which is the paraphrase word of “Hokkaido University”, is not registered in the speech recognition target vocabulary database of the external storage device 15, the speech recognition device can recognize even if the user utters “Hokkaido University”. , Voice recognition results with “Hokkaido University” cannot be output.

  On the other hand, the speech recognition apparatus to which the present invention is applied evaluates the actual usage of the paraphrase actually used by the user. For example, when the actual usage of the paraphrase “Hokkaido” is high, The paraphrase word “Hokkaido University” is registered in the speech recognition target vocabulary database. That is, as shown in FIG. 5, the voice recognition device emits a notification voice from the speaker 4 that the user speaks “destination” in (1) and “please give a destination command” in (2). In (3), the user utters “facility”, and in (4), when the user utters “Hokkaido Univ.” Since the speech recognition target vocabulary of the “Hokkaido University” is registered in the speech recognition target vocabulary database as the paraphrase of the official name “Hokkaido University”, sound emission from “Hokkaido University” and the speaker 4 in FIG. In the speech recognition result image 44 shown in d), “Hokkaido University” can be displayed in the speech recognition result display field 45.

  Hereinafter, a process for registering a paraphrase word having a high actual use degree in the speech recognition target vocabulary database will be described with reference to FIG.

  First, in step S1, the CPU 21 of the signal processing device 11 detects that the user has operated the speech switch 5a, determines that the start of speech has been instructed, and advances the processing to step S2.

  In step S2, the CPU 21 performs standby setting for voice recognition processing. The CPU 21 displays one of the images shown in FIGS. 3A to 3D as a voice input menu and enters a voice input standby state. Immediately after the utterance switch 5a is operated in step S1, the voice menu image 41 of FIG. 3A, which is the highest hierarchy, is displayed and enters a standby state. For example, “destination”, “facility name” The operation names are selected and the images shown in FIGS. 3B to 3D are displayed as the operation name is moved to the lower layer.

  Further, the CPU 21 reads the speech recognition target vocabulary registered in the speech recognition target vocabulary database from the external storage device 15 into the memory 22. Then, with the user's utterance, the hierarchy advances to a lower layer such as the operation name databases 31 and 32, and the vocabulary registered in the speech recognition target vocabulary database 33 in FIG. 2, that is, the facility name is read as the speech recognition target vocabulary. Suppose that In addition, it is difficult to read all facility names in the country into the memory 22 for reasons such as an increase in memory capacity and voice recognition calculation speed, but facilities in the vicinity of the user's location and typical facilities nationwide that have been determined in advance. Suppose the name has been read.

  In the next step S3, the CPU 21 sends a notification voice signal stored in the external storage device 15 to the D / A converter 13 and the amplifier 14 in order to notify the user that the voice recognition process has been started. To the speaker 4 and the announcement sound is emitted from the speaker 4. For example, when a facility name is uttered, “Please name the facility” corresponds to the announcement voice.

  For example, in the scene where the facility name input image 43 shown in FIG. 3C is displayed in response to the announcement voice, the paraphrase “Hokkaido University” of the official name “Hokkaido University” as shown in FIG. Suppose you speak. In this case, when the signal processing unit 1 converts the audio signal from the microphone 2 into a digital signal by the A / D converter 12 and inputs the digital signal to the signal processing device 11, the signal processing unit 1 is digital until the speech switch 5 a is operated by the CPU 21. The average power of the signal is calculated. After the utterance switch 5a is operated, the signal processing device 11 determines that the user has uttered when the instantaneous power of the digital signal is greater than a predetermined value as compared with the average power, and starts to capture voice. .

  In the next step S <b> 4, the CPU 21 starts reading the audio signal in step S <b> 3 and starts calculating the degree of coincidence between the read audio signal and the speech recognition target vocabulary stored in the memory 22. The degree of coincidence, that is, the degree of coincidence between the speech segments and the speech signals indicating the individual speech recognition target words, is obtained as a score for each speech segment by the CPU 21. It is assumed that the greater the score, the higher the degree of matching. Note that voice capturing is continued in parallel with the processing for obtaining the degree of coincidence for each voice section.

  In the next step S5, the CPU 21 determines that the user's utterance has ended when the instantaneous power of the digital audio signal obtained from the A / D converter 12 has fallen below a predetermined value over a predetermined time. , End audio capture.

  In the next step S6, the CPU 21 determines whether or not the predetermined name input hierarchy is displaying the voice recognition result display field 45 as shown in FIG. That is, the destination facility is not a hierarchy in which operation names such as a destination and a search condition are input like the hierarchy in which the voice menu image 41 and the destination selection menu image 42 are displayed as shown in FIGS. It is determined whether or not it is a hierarchy in which a formal name such as a name and address and a paraphrase of the formal name are input. For example, when the hierarchy advances with the utterance and the hierarchy at the time when the voice capturing is completed is the hierarchy where the facility name input image 43 of FIG. 3C is displayed, there is a possibility that the formal name and the paraphrase are uttered. It is determined that there is a certain hierarchy, and the process proceeds to step S7. On the other hand, when it is determined that the operation name to be selected from the menu list 46 is determined as in the voice menu image 41 or the destination selection menu image 42, the process proceeds from step S6 to step S8.

  In step S7, the CPU 21 considers the possibility that the user's utterance “Hokkaido University” is a paraphrase word in the subsequent step S14, and outputs a digital signal indicating the voice of “Hokkaido University” that has finished the voice capture in step S5. Temporarily stored in the memory 22.

  In the next step S8, the CPU 21 obtains the degree of coincidence between the speech recognition target vocabulary stored in the speech recognition target vocabulary database and the digital signal indicating the speech of “Hokkaido University”, and the speech recognition results in descending order of coincidence. Get candidates for.

  In the next step S9, the CPU 21 outputs the speech recognition result candidate acquired in step S8. For example, as shown in FIG. 3 (d), the CPU 21 displays in the speech recognition result display field 45 of the speech recognition result image 44 the speech recognition target vocabulary “Kokubun Station” having the highest degree of coincidence with the digital signal indicating “Hokkaido University” speech. Is output as a speech recognition result. As a method of outputting the voice recognition result, “Kokubun Station” of the voice recognition result is converted into a voice signal by the voice synthesizing function of the CPU 21, and “D / A converter 13 and the amplifier 14 are used to output“ Kokubun Station "may be sounded.

  Thereafter, the correction switch 5b is operated by the user in response to outputting “Kokubun Station” as the voice recognition result. As a result, in step S10, the signal processing device 11 determines whether or not it has been detected that the correction switch 5b has been operated within a predetermined time (for example, several tens of seconds) after outputting the speech recognition result in step S9. To do. If it is detected that the correction switch 5b has been operated within a predetermined time, the process proceeds from step S10 to step S11. If it is not detected that the correction switch 5b has been operated within the predetermined time, the process proceeds to step S10. To step S12.

  In step S11, the CPU 21 increments and records the number of operations of the correction switch 5b, advances the process to step S3, and repeats the processes of steps S3 to S10. After that, the utterance “Hokkaido University” was repeated from the user as shown in (9) to (16) of FIG. 4, and then the official name “Hokkaido University” was uttered in (17) of FIG. And In this case, since the official name “Hokkaido University” is registered in the speech recognition target vocabulary database 33 in step S8, the speech recognition target vocabulary “Hokkaido University” has the highest degree of coincidence, and in step S9. The speech recognition result “Hokkaido University” can be output.

  Thus, in step S10 after outputting the speech recognition result “Hokkaido University”, the correction switch 5b is not operated, and the CPU 21 advances the process from step S10 to step S12.

  In step S12, the CPU 21 determines whether the speech recognition target vocabulary for which the speech recognition result is output in step S9 is an operation name, a formal name, or a paraphrase, and determines whether there is a next hierarchy. To do. If there is a next hierarchy, the process returns to step S2. If there is no next hierarchy, the process proceeds to step S13. For example, when “Hokkaido University” of the facility name is output as the speech recognition result as the destination, the process proceeds to step S13.

  In step S <b> 13, the CPU 21 determines the speech recognition result when the “go there” button 47 or the “view map” button 48 included in the speech recognition result image 44 of FIG. 3D is selected. When the “go to there” button 47 or the “view map” button 48 is selected, the location information and commands of Hokkaido University are supplied to a navigation device (not shown) to search for a route or display a map.

  In the next step S <b> 14, the CPU 21 evaluates the possibility that “Hokkaido University” uttered by the user is a paraphrase of the official name “Hokkaido University”.

  First, the CPU 21 generates a paraphrase from the official name “Hokkaido University”. The CPU 21 divides the official name into morphemes by a morpheme analysis program, and divides them into “Hokkaido” and “University”. The morphological analysis is realized by executing a general-purpose program (for example, ChaSen-http: //Chasen.aist-nara.ac.jp/) on the CPU 21. The CPU 21 generates a plurality of paraphrases by taking out and concatenating partial character strings from the two divided morphemes “Hokkaido” and “University”. For example, two-letter paraphrases are “Hokkaido University”, “Umi Univ.”, “Michidai”, “Hokkaido”, “Oceanology”, “Dogaku”, and other paraphrasing numbers are also generated. The

  Next, the CPU 21 evaluates the actual usage of the paraphrase words spoken by the user and temporarily stored in the memory 22 in step S7. At this time, the CPU 21 operates the correction switch 5b in step S10, and the greater the number of operations of the correction switch 5b in step S11, the higher the user's intention to use the paraphrase, and the actual usage of the paraphrase is higher. Evaluate as high. Then, the CPU 21 determines whether or not the number of operations of the correction switch 5b is equal to or greater than a predetermined value. If the number of operations of the correction switch 5b is equal to or greater than the predetermined value, the actual usage of the paraphrase is high. Decide to register in the database.

  The user's intention to use is not limited to the number of operations of the correction switch 5b, but may be the number of times the same paraphrase “Hokkaido” is spoken. For example, if the paraphrase words “Hokkaido University” and “Hokkaido University” for the official name “Hokkaido University” are temporarily stored in the memory 22 in step S7, and the paraphrase word “Hokkaido University” is spoken more frequently, Can be determined to have a higher intended use.

  Next, the CPU 21 uses the user's speech “Hokkaido University” temporarily stored in the memory 22 in step S7 as the input speech, and recognizes all of the paraphrases generated from the two morphemes “Hokkaido” and “University” as speech recognition targets. The vocabulary is used, and the degree of coincidence between the input speech and the speech recognition target vocabulary is calculated. As a result, when there is a paraphrase having a matching score equal to or higher than a predetermined threshold, the CPU 21 registers the input speech as a speech recognition target vocabulary in the speech recognition target vocabulary database.

  Further, the CPU 21 may generate and register a similar paraphrase for another university name in the same category as “Hokkaido University”. That is, when “Hokkaido University” is registered as a speech recognition target vocabulary in the speech recognition target vocabulary database, the CPU 21 connects the first characters of “Hokkaido” and “University”, which are the results of morphological analysis, and rephrases “Hokkaido University”. A rule for creating a word may be generated, and the rule may be applied to a formal name of another university and registered as a speech recognition target vocabulary.

  Thus, after registering the official name “Hokkaido University” and the paraphrase word “Hokkaido University” in the speech recognition target vocabulary database, as shown in FIG. 5 (5), the response to the user speaking “Hokkaido University” The official name “Hokkaido University” can be output as a speech recognition result. In addition, the CPU 21 outputs a speech recognition result with “Hokkaido University” as a response to what the user spoke with “Hokkaido University”, and at the same time, speaks “Hokkaido University” as the paraphrase for the official name “Hokkaido University”. However, the fact that the official name “Hokkaido University” can be output as the voice recognition result may be displayed in the voice recognition result image 44 of FIG.

[Effect of the first embodiment]
As described above in detail, according to the speech recognition apparatus according to the first embodiment to which the present invention is applied, a paraphrase with high actual usage is registered as a speech recognition target vocabulary, so that it may be actually used. Only high paraphrasing words can be additionally registered as speech recognition target vocabulary, and there is no problem that the number of speech recognition target vocabularies increases more than necessary, and the usability of voice operation can be greatly improved.

  In addition, according to this speech recognition apparatus, the paraphrase (first speech recognition result) corrected by the correction switch 5b is stored, and then the speech recognition result (second speech recognition result) based on the input speech. ) Is not corrected, it is evaluated that the corrected speech recognition result (first speech recognition result) is highly used as a paraphrase for the uncorrected speech recognition result (second speech recognition result). Therefore, only the corrected paraphrase word can be registered as the speech recognition target vocabulary, and it is possible to avoid increasing the speech recognition target vocabulary more than necessary.

  Furthermore, according to this speech recognition apparatus, since the paraphrase word is registered when the intended use of the paraphrase word is high from the number of times the correction switch 5b has been operated, the operation is led to input by using many operations. Only paraphrases with high intent can be additionally registered in the speech recognition result, and an increase in the speech recognition result can be avoided.

  Furthermore, according to this speech recognition apparatus, when a paraphrase word is registered as a speech recognition target vocabulary, a rule for generating the paraphrase word from the official name is obtained and classified into the same category as the paraphrase word. Since the official name “Hokkaido University” is registered as “Hokkaido University” in a category such as a university, for example, the official name “ABC University” to “A University” in the same category is registered. Can be registered. This makes it easy to use the speech recognition apparatus using paraphrased words.

  Furthermore, according to the speech recognition apparatus, since the paraphrase registered as the speech recognition target vocabulary is presented to the user, it is possible to promptly use the paraphrase from the next use. It becomes.

[Second Embodiment]
Next, a speech recognition apparatus according to the second embodiment will be described. In addition, since the structure of the speech recognition apparatus according to the second embodiment is the same as that of the above-described first embodiment, detailed description thereof is omitted by attaching the same reference numerals.

  When the user wants to go to a place with the official name “Atsugi International Country Club” or display a map, the paraphrase “Atsugi Country” is not registered in the speech recognition target vocabulary database. The operation will be described.

  In such a voice recognition device, voice input is accepted when the voice menu image 41 for voice input shown in FIG.

  In a state where the voice menu image 41 is displayed, when the user speaks “destination” as shown in (1) of FIG. 7 and obtains a voice recognition result of the “destination”, the CPU 21 7, a notification voice saying “Please give a destination command” is emitted from the speaker 4, and a destination selection menu including a menu list 46 showing a plurality of operation names in FIG. The image 42 is displayed on the touch panel display 3.

  In the state where the destination selection menu image 42 of FIG. 3B is displayed, the user selects the operation name “facility” included in the menu list 46 of the destination selection menu image 42 as shown in FIG. When the speech recognition result of “facility” is obtained by the utterance and the CPU 21, an announcement voice “Please name the facility” is emitted from the speaker 4 as shown in FIG. The facility name input image 43 shown in (c) is displayed.

  When the user utters “Atsugi Country” as shown in (5) of FIG. 7 while the facility name input image 43 of FIG. 3C is displayed, the CPU 21 recognizes the voice of the external storage device 15. Since “Atsugi Country” is not registered as a speech recognition target vocabulary in the target vocabulary database, select the Atsugi station registered in the speech recognition target vocabulary database closest to the speech recognition result for “Atsugi Country”. Then, the voice recognition result image 44 including the voice recognition result “Atsugi Station” is displayed in the voice recognition result display field 45 of FIG.

  When the user operates the correction switch 5b as shown in (7) of FIG. 7 in the state where the voice recognition result image 44 is displayed, “speak again” from the speaker 4 as shown in (8) of FIG. The notification voice “Please” is emitted and the facility name input image 43 of FIG. 3C is displayed again. If the user again utters “Atsugi Country” and outputs the speech recognition result “Atsugi Station” in (10) of FIG. 7, the user gives up the input by voice, 7 (11), “Atsugi International Country Club” is input by manual operation using the touch panel display 3.

  As described above, the speech recognition apparatus according to the second embodiment includes the alternative input unit that replaces the speech recognition, and when the formal input different from the speech recognition result is input by the alternative input unit, the speech recognition The result is evaluated as having high actual usage as a paraphrase of the formal name input by the alternative input means, and is registered as a speech recognition target vocabulary.

  Then, the voice recognition device utters “Destination” in (1) in FIG. 8 and emits a notification voice from the speaker 4 that “Please give a destination command” in (2). (3) When the user utters “facility” in (4) and utters an announcement sound “Please name the facility” from the speaker 4, the user utters “Atsugi country”. Since the speech recognition result of “Country” is registered in the speech recognition target vocabulary database, sound is output from “Atsugi International Country Club” and the speaker 4 in FIG. 8 (6), and the speech recognition result image shown in FIG. 44, “Atsugi International Country Club” can be displayed in the voice recognition result display field 45.

  Hereinafter, the operation of the speech recognition apparatus according to the second embodiment will be described with reference to FIGS. 9 and 10.

  As shown in FIG. 9, the speech recognition apparatus according to the second embodiment displays a menu list 46 including operation names as shown in FIGS. 3 (a) and 3 (b) and selects an operation as shown in FIG. c) A process for obtaining a speech recognition result of a formal name or paraphrase as in (d) is performed. Similar to the speech recognition device of the first embodiment, the speech recognition device performs the processing of step S1 to step S9, and when it is detected in step S10 that the correction switch 5b has been operated within a predetermined time, The process returns to step S3, and if the correction switch 5b is not operated within a predetermined time, the process proceeds to step S12. As described above, since the speech recognition apparatus according to the second embodiment evaluates that the actual usage of the paraphrase word is high when the formal name is input by the alternative input unit, as shown in step S11 of FIG. The process of recording the number of operations of the correction switch 5b is not performed. Then, in step S13 after determining that there is no next lower layer in step S12, the speech recognition result output in step S9 is determined, and the process ends.

  Here, as described above, when the user gives up the voice input of the paraphrase word “Atsugi Country” and causes the voice recognition device to recognize the official name “Atsugi International Country Club” by the operation input by the touch panel display 3, the voice recognition is performed. The apparatus registers a paraphrase word having a high actual use degree in the speech recognition target vocabulary database by performing the processing shown in FIG.

  As shown in FIG. 10, when the CPU 21 first detects in step S21 that a menu switch of the input device 5 (not shown) has been operated, it proceeds to step S22 and follows the operation of the menu switch. When the menu screen is set and displayed, and the operation input by the user is confirmed in order to make a transition to the screen for inputting the formal name by the touch panel display 3 in step S23, the lower layer of the currently displayed screen is displayed in step S24. Determine if it exists.

  If the CPU 21 determines in step S24 that there is no lower layer screen, such as an alternative input screen 51 for inputting a formal name instead of voice input, as shown in FIG. 11, the process proceeds to step S25. To proceed. The alternative input screen 51 includes category information 52 of a formal name to be input by the user, a formal name input field 53, a 50-sound character input button 54, and a list display button 55. The category information 52 indicates the facility that is the category selected by the user in step S22 and step S23, and the golf course that is the lower-level category of the facility.

  In step S25, the CPU 21 detects that the character input button 54 and the list display button 55 on the alternative input screen 51 are operated by the user, and performs a process of determining the operation result. At this time, as shown in FIG. 11, when a part of the official name “Atsugi International Country Club” of the facility is input by operating the character input button 54 and then the list display button 55 is operated. As shown in FIG. 12, a list display screen 61 in which the speech recognition target vocabulary including “Ajito” at the head is listed is displayed. At this time, the CPU 21 performs a partial match search including “Aki” in the speech recognition target vocabulary whose category is the facility, and the speech recognition target including “Aso” from the speech recognition target vocabulary database of the external storage device 15. Extract vocabulary. The list display screen 61 includes a list display 62 including a search key “Aki”, a “go there” button 63 and a “view map” button 64.

  After the list display screen 61 is displayed, when “Atsugi International Country Club” is selected from the list display 62 by the user, the CPU 21 detects the operation and determines the operation result. In addition, when the user selects “Atsugi International Country Club” and further selects the “go there” button 63 or the “view map” button 64, the operation content may be determined.

  Next, in step S <b> 26, the CPU 21 determines whether or not the process of inputting the voice in FIG. 9 has been performed in the time zone (for example, several minutes) immediately before the operation result is determined in step S <b> 25. At this time, for example, if the CPU 21 is configured to hold the audio digital signal temporarily stored in the memory 22 for a predetermined period, the audio digital signal is temporarily stored in the memory 22 in step S7 of FIG. When it is determined that there is a voice input, it is determined that there was a voice input immediately before.

  In the next step S27, as determined in step S26, the CPU 21 generates a paraphrase from the voice input immediately before the transition from the alternative input screen 51 to the list display screen 61 and the official name is selected, Processing to register in the speech recognition target vocabulary database is performed. For example, since the user does not know the official name, there is a possibility that the voice recognition apparatus cannot recognize the official name by voice input, and the process is interrupted and the official name may be input from the alternative input screen 51. Process to generate words.

  In step S27, the CPU 21 reads out from the memory 22 a voice digital signal related to the voice input performed immediately before, and generates a paraphrase from the voice digital signal. Next, the CPU 21 compares the generated paraphrase with the digital audio signal stored in the memory 22 and determines that the paraphrase having a high degree of coincidence is a paraphrase for the formal name. At this time, the CPU 21 creates a candidate for the paraphrase “Atsugi Country” by combining the morphemes “Atsugi”, “International”, “Country”, “Club” from the official name “Atsugi International Country Club”, for example. When “Atsugi Country” is stored in the memory 22, it is determined that the “Atsugi Country” is a paraphrase of “Atsugi International Country Club”, and is registered in the speech recognition target vocabulary database.

  In addition, the digital signal of the voice of the initial utterance of the user is selected from the digital signals of the voice stored in the memory 22 and collated with the paraphrase generated from the official name, and the digital signal of the voice of the initial utterance is When the likelihood of a paraphrase word is high, it is desirable to register the paraphrase word in the speech recognition target vocabulary database.

[Effects of Second Embodiment]
As described above in detail, according to the speech recognition apparatus according to the second embodiment to which the present invention is applied, when an official name is input on the alternative input screen 51, speech input is performed before the official name is input. If there is a word, the paraphrase word of the speech can be registered in the speech recognition target vocabulary database as a paraphrase word having a high actual usage rate. Can be registered in the speech recognition target vocabulary database, and the number of speech recognition target vocabulary can be prevented from being increased more than necessary.

  Further, according to this speech recognition apparatus, when a formal name is input on the alternative input screen 51, the paraphrase generated from the formal name is collated with the voice of the user's initial utterance stored in the memory 22. Then, when the likelihood is high, the actual usage of the paraphrase word is evaluated and registered, so that the user's unique paraphrase word can be registered in the speech recognition target vocabulary database, and the speech recognition target vocabulary is more than necessary. Can be avoided.

[Third Embodiment]
Next, a speech recognition apparatus according to the third embodiment will be described. Note that parts similar to those in the above-described embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.

  As shown in FIG. 13, the speech recognition apparatus according to the third embodiment includes a communication device (communication device) connected to an information content storage server (information content storage means, not shown) via a network. Means) 70 is connected to the speech recognition apparatus according to the above-described embodiment. The communication device 70 performs communication processing according to a communication protocol such as IP (Internet Protocol), for example, in accordance with an instruction from the signal processing unit 1.

  For example, if the official name of the destination facility name is “Kwansei Gakuin University” and the paraphrase word “Kangaku” is not registered in the speech recognition target vocabulary database, this speech recognition apparatus has a name as shown in FIG. When the paraphrase word “Kakaku” is registered in the speech recognition target vocabulary database as described later, the processing shown in FIG. 15 is performed.

  When the user utters “destination” as shown in (1) of FIG. 14 and obtains a voice recognition result of the “destination” as shown in (1) of FIG. Announcement voice “Please command” is emitted from speaker 4. Next, as shown in FIG. 14 (3), when the user speaks the operation name “facility” and the signal processing unit 1 obtains a speech recognition result of “facility”, the operation (4) in FIG. As shown, the announcement sound “Please name the facility” is emitted from the speaker 4.

  Next, when the user utters “Sekigaku” as shown in FIG. 14 (5), the signal processing unit 1 has “Sekigaku” as a speech recognition target vocabulary in the speech recognition target vocabulary database of the external storage device 15. Since it is not registered, the Kofu station registered in the speech recognition target vocabulary database closest to the speech recognition result for the “Sekigaku” is selected. Next, when the user operates the correction switch 5b as shown in FIG. 14 (7), the signal processing unit 1 notifies the speaker 4 “Please speak again” as shown in FIG. 14 (8). When the voice is emitted and the user speaks “Sekigaku” again, when the voice recognition result “Kofu Station” is output in (10) of FIG. 14, the user gives up the voice input. Then, in (11) of FIG. 14, “Kwansei Gakuin University” is input by manual operation using the touch panel display 3.

  In (11) of FIG. 14, the signal processing unit 1 operates the character input button 54 on the alternative input screen 51 of the country “University” as shown in FIG. When the list display button 55 is further operated, a list display screen 61 shown in FIG. 17 is displayed. Then, “Kwansei Gakuin University” is selected from the list display 62 of the list display screen 61, and the official name “Kwansei Gakuin University” is input.

  Next, the signal processing unit 1 generates the paraphrase word “Kangaku” from the official name “Kwansei Gakuin University” and stores the generated paraphrase word “Kangaku” as a search key in the information content storage server on the network. The communication device 70 is controlled to search for information content. Then, when the generated paraphrase word “Sekigaku” is included in the information content stored in the information content storage server connected by the communication device 70, the signal processing unit 1 executes the actual wording of the paraphrase word “Sekigaku”. Assess that the usage is high and register it in the speech recognition target vocabulary database.

  Then, the voice recognition device utters “Destination” in (1) in FIG. 15, and emits a notification voice from the speaker 4 that “Please give a destination command” in (2). (3) When the user utters “facilities” in (4) and utters the announcement sound from the speaker 4 that “please name the facility”, the user utters “Sekigaku”. 15 is registered in the speech recognition target vocabulary database, so that “Kwansei Gakuin University” and sound are emitted from the speaker 4 in (6) of FIG. 15 and the speech recognition result image 44 shown in FIG. “Kwansei Gakuin University” can be displayed in the recognition result display field 45.

  As shown in FIG. 18, the voice recognition apparatus performs the steps S8 to S10, S12, and S13 after capturing the voice uttered by the user in the processes of steps S1 to S5. Here, in contrast to the process of FIG. 9 performed by the speech recognition apparatus according to the second embodiment, the processes of step S6 and step S7 are not performed by the speech recognition apparatus according to the third embodiment. The reason for this is that the speech recognition apparatus according to the third embodiment uses the alternative input screen 51 and the list display screen 61 to generate a paraphrase having a high actual usage rate from the formal name input. This is because step S7 for storing the voice of the paraphrase word spoken by the person in the memory 22 is not performed.

  In addition, the speech recognition apparatus according to the third embodiment performs the processing of step S21 to step S26 as in FIG. 10 described in the second embodiment, and in step S26, the processing illustrated in FIG. 18 is performed immediately before. If it is determined that the paraphrase word is generated in step S27, a process of registering the paraphrase word having a high actual usage in the speech recognition target vocabulary database is performed.

  In step S27, the signal processing unit 1 first generates a paraphrase from the official name “Kwansei Gakuin University”. At this time, the signal processing unit 1 divides the official name “Kwansei Gakuin University” into morphemes by the morphological analysis program, and divides them into “Kansai”, “Gakuin”, and “University”. Next, the signal processing unit 1 extracts partial character strings from the three morphemes and connects them to generate a plurality of paraphrases. For example, the two-letter paraphrases are "Sekigaku", "Nishigaku", "Sekiin", "Saiin", "Sekidai", "Nishidai", "Gakudai", "Shondai", "Sekigaku", "Nishigaku" ”,“ Study ”, and“ Study ”, and generate other paraphrasing words.

  Next, the signal processing unit 1 controls the communication device 70 to access information content on the network and evaluates whether the paraphrase is actually used. At this time, the signal processing unit 1 first causes the communication device 70 to search for information content such as a Web homepage using the generated paraphrase as a search key. Next, the signal processing unit 1 acquires the number of search results from the communication device 70, determines whether or not the number of search results is equal to or greater than a predetermined threshold, and if the number is greater than or equal to the threshold, information content of the search result It is determined whether there is information content having a formal name. When the formal name exists in the information content, the signal processing unit 1 determines that the paraphrase used as the search key is actually used, and registers the paraphrase as a speech recognition target vocabulary.

  As described above, as a result of searching the information content by using “Kan Gaku” among the paraphrases generated from the official name “Kwansei Gakuin University” by the signal processing unit 1 as a search key, the number of information contents as the search result is a predetermined number. The paraphrase with high actual usage as described above, and when the information content includes the official name “Kwansei Gakuin University”, the paraphrase “Kangaku” can be registered as a speech recognition target vocabulary.

  In addition, when the information content obtained as a result of searching the search key as a paraphrase by the communication device 70 is described in a structured language such as HTML (Hypertext Markup Language), the signal processing unit 1 Judge whether the official name exists in the title part. If the formal name exists in the title portion of the HTML data, it may be evaluated that the actual usage of the paraphrase word as the search key is high and registered as a speech recognition target vocabulary.

  Further, when the category of the formal name obtained from the alternative input screen 51 and the list display screen 61 is the location name, the signal processing unit 1 uses the location name that is the official name as a search key for retrieving information content by the communication device 70. It is desirable to add not only the name but also the location information of the point name.

[Effect of the third embodiment]
As described above in detail, according to the speech recognition apparatus according to the third embodiment to which the present invention is applied, when a paraphrase searched by the communication device 70 is included in a plurality of information contents, the paraphrase It is possible to register paraphrasing words that are more commonly used than those that were not recognized at the time of design, so that the recognition rate of paraphrasing words can be increased and necessary. As described above, an increase in the number of speech recognition target words can be avoided.

  Further, according to the speech recognition apparatus, when the number of information contents including a paraphrase word is equal to or greater than a predetermined value, it is evaluated that the actual usage of the paraphrase word is high. By increasing the predetermined value, the accuracy of registering paraphrased words can be improved, the ease of use can be greatly improved, and an increase in the vocabulary for speech recognition can be avoided.

  Furthermore, according to the speech recognition apparatus, when both the formal name and the paraphrase generated from the formal name co-occur in the information content retrieved by the communication device 70, the paraphrase word Since it is evaluated that the actual usage is high, the accuracy of registering paraphrased words can be greatly improved, and the usability can be greatly improved.

  Furthermore, according to the speech recognition apparatus, when the information content is searched using the paraphrase word as a search condition and the official name is included in the searched information content, it is evaluated that the actual usage of the paraphrase word is high. Even personal information content can be obtained as a search result, a new paraphrase can be registered more quickly, and the usability can be greatly improved.

  Furthermore, according to the speech recognition apparatus, when the information content is searched using the paraphrase word as a search condition, and the official name is included in the title portion of the information content, it is evaluated that the actual usage of the paraphrase word is high. The accuracy of registering paraphrased words can be made extremely high.

  Furthermore, according to the speech recognition apparatus, when the category of the official name is a spot name, the search is performed by including the information content of the spot name in the information content search condition. Avoid registering words.

  The above-described embodiment is an example of the present invention. For this reason, the present invention is not limited to the above-described embodiment, and various modifications can be made depending on the design and the like as long as the technical idea according to the present invention is not deviated from this embodiment. Of course, it is possible to change.

It is a block diagram which shows the structure of the speech recognition apparatus which concerns on 1st Embodiment to which this invention is applied. It is a figure which shows the operation name database and speech recognition object vocabulary database in the speech recognition apparatus which concerns on 1st Embodiment to which this invention is applied. It is a figure explaining the screen transition in the speech recognition device concerning a 1st embodiment to which the present invention is applied, (a) is a voice menu image, (b) is a destination selection menu image, (c) is a facility name input image, (D) is a speech recognition result image. It is a figure which shows operation | movement when the paraphrase word is not registered in the speech recognition apparatus which concerns on 1st Embodiment to which this invention is applied. It is a figure which shows operation | movement when the paraphrase word is registered in the speech recognition apparatus which concerns on 1st Embodiment to which this invention is applied. It is a flowchart which shows the process sequence of the speech recognition apparatus which concerns on 1st Embodiment to which this invention is applied. It is a figure which shows operation | movement when the paraphrase word is not registered in the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a figure which shows operation | movement when the paraphrase word is registered in the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a flowchart which shows the process sequence at the time of the speech recognition by the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a flowchart which shows the process sequence at the time of registration of the paraphrase word by the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a figure which shows the alternative input screen in the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a figure which shows the list display screen in the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a block diagram which shows the structure of the speech recognition apparatus which concerns on 3rd Embodiment to which this invention is applied. It is a figure which shows operation | movement when the paraphrase word is not registered in the speech recognition apparatus which concerns on 3rd Embodiment to which this invention is applied. It is a figure which shows operation | movement when the paraphrase word is registered in the speech recognition apparatus which concerns on 3rd Embodiment to which this invention is applied. It is a figure which shows the alternative input screen in the speech recognition apparatus which concerns on 3rd Embodiment to which this invention is applied. It is a figure which shows the list display screen in the speech recognition apparatus which concerns on 2nd Embodiment to which this invention is applied. It is a flowchart which shows the process sequence at the time of the speech recognition by the speech recognition apparatus which concerns on 3rd Embodiment to which this invention is applied.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Signal processing unit 2 Microphone 3 Touch panel display 4 Speaker 5 Input device 5a Speech switch 5b Correction switch 11 Signal processing device 12 A / D converter 13 D / A converter 14 Amplifier 15 External storage device 21 CPU
22 Memory 31, 32 Operation name database 33 Voice recognition target vocabulary database 41 Voice menu image 42 Pre-selected menu image 43 Facility name input image 44 Voice recognition result image 45 Voice recognition result display column 46 Menu list 47, 63 “Go there” Buttons 48 and 64 “View Map” Button 51 Alternative Input Screen 52 Category Information 53 Formal Name Input Field 54 Character Input Button 55 List Display Button 61 List Display Screen 62 List Display 70 Communication Device

Claims (13)

A speech recognition device comprising speech recognition means for recognizing speech emitted from a user,
Storage means for storing the official name as a speech recognition target vocabulary;
Paraphrase word generating means for generating a paraphrase word from the formal name stored in the storage means;
Actual usage evaluation means for evaluating the actual usage of the paraphrase generated by the paraphrase generation means;
Registration means for registering the paraphrase word generated by the paraphrase word generation means in the storage means as a speech recognition target vocabulary;
The said registration means registers only the paraphrase word evaluated that the actual usage is high by the said practicality evaluation means as a speech recognition object vocabulary, The speech recognition apparatus characterized by the above-mentioned. A correction instruction means for inputting an instruction to correct the voice recognition result generated by the voice recognition means for the voice emitted from the user;
The actual usage evaluation means temporarily stores the first voice recognition result when an instruction to correct the first voice recognition result is input by the correction instruction means, and then the voice recognition means When the generated second speech recognition result is determined without being corrected by the correction instruction means, the first speech recognition result is used as a paraphrase for the second speech recognition result. The speech recognition apparatus according to claim 1, wherein the speech recognition apparatus is evaluated as being high. A use intention judging means for judging the use intention of the paraphrase word from the operation amount of the correction instruction means of the user;
The speech recognition apparatus according to claim 2, wherein the actual usage evaluation unit evaluates a paraphrase that has been determined to have a high use intention by the use intention determination unit as a paraphrase having a high actual usage.   When the paraphrase word is registered, the registration means obtains a rule in which the paraphrase word is generated from the formal name, and converts the paraphrase word of another formal name classified into the same category as the paraphrase word. The speech recognition apparatus according to claim 1, wherein registration is performed according to the following. It further comprises an alternative input means for inputting a formal name by a user operation instead of voice,
When the actual usage evaluation means inputs a formal name different from the voice recognition result of the voice recognition means by the alternative input means, the voice recognition result of the voice recognition means is input by the alternative input means. The speech recognition apparatus according to claim 1, wherein the speech recognition apparatus evaluates that the actual usage is high as a paraphrase of the official name.   The actual usage evaluation means collates the initial utterance of the user with the paraphrase generated by the paraphrase generating means in the speech recognition result of the speech recognition means, and the user's initial utterance and the paraphrase having a high likelihood. The speech recognition apparatus according to claim 1, wherein the speech recognition apparatus evaluates that the word usage is high. A communication means for searching for information content included in the network;
If the paraphrase word is included in a plurality of information contents as a result of searching the information content for the paraphrase word generated by the paraphrase word generation unit by the communication unit, The speech recognition apparatus according to claim 1, wherein the speech recognition apparatus evaluates that the word usage is high.   The actual usage evaluation unit determines that the actual usage of the paraphrase word is high when the number of information contents including the paraphrase word generated by the paraphrase word generation unit is equal to or greater than a predetermined value. The speech recognition apparatus according to claim 7, wherein   The actual usage evaluation means, when both the formal name and the paraphrase generated by the paraphrase generation means from the official name co-occur in the information content, the actual use of the paraphrase word The speech recognition apparatus according to claim 7, wherein the speech recognition apparatus evaluates that the degree is high.   The actual usage evaluation unit causes the information content to be searched by using the paraphrase word generated by the paraphrase word generation unit as a search condition by the communication unit, and the formal name is included in the information content searched by the communication unit The speech recognition apparatus according to claim 7, wherein, when it is determined, the actual usage of the paraphrase word is evaluated as being high.   The actual usage evaluation unit causes the information content to be searched by using the paraphrase word generated by the paraphrase word generation unit by the communication unit as a search condition, and the title part of the information content searched by the communication unit is officially displayed. The speech recognition apparatus according to claim 7, wherein when the name is included, it is determined that the actual usage of the paraphrase is high.   8. The speech recognition apparatus according to claim 7, wherein, when the category of the official name is a spot name, the communication unit performs a search by including position information of the spot name in the search condition of the information content. .   The speech recognition apparatus according to claim 1, wherein the paraphrase word registered by the registration unit is presented to the user that it can be used.

Images