JP2010054896A - Voice recognition device and voice recognition program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition device and a voice recognition program, for recognizing voice with high accuracy by using a word dictionary. <P>SOLUTION: When a visitor is detected, voice for a question about his or her name and the name of a person in charge is output from a reception device(S42). Voice of an answer to the question from the visitor is recognized by referring to a general purpose dictionary which shows correspondence between a word which belongs to a plurality of categories and information on its pronunciation, and a division dictionary which is created from the general purpose dictionary based on categories for each person in charge. A final result is obtained which is determined based on a recognition result obtained by referring to the general purpose dictionary, and a recognition result obtained by referring to the division dictionary (S44). When the final result indicates recognition failure (S45:YES), retry is requested to the visitor (S47). When the recognition is successful (S45:NO), the name of the person in charge included in the final result is specified (S52), his or her contact information is specified (S53), and notification is made(S54). <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus and a speech recognition program. More specifically, the present invention relates to a speech recognition apparatus and speech recognition program that perform speech recognition using a word dictionary.

  Conventionally, speech recognition technology using an acoustic model that shows the acoustic characteristics of speech, a word dictionary that describes the correspondence between a word and information related to the pronunciation of the word, and a language model that defines constraints on word connections has been known. Yes. In such a technique, as the number of words included in the word dictionary increases, the number of sentences composed of word combinations becomes enormous. As a result, the probability of misrecognition in which a sentence different from that actually spoken is output tends to increase, that is, the recognition accuracy tends to decrease.

Therefore, a speech recognition device that suppresses misrecognition by recognizing only a sentence generated only by a predicted word as a recognition candidate has been proposed (for example, Patent Document 1).
JP-A-6-180593

  However, in the speech recognition apparatus described in Patent Document 1, only predicted words are used for speech recognition. Therefore, when a sentence actually spoken is not composed of predicted words, that is, prediction is performed. If it falls off, recognition may not be possible.

  The present invention has been made to solve the above problems, and an object of the present invention is to provide a speech recognition apparatus and a speech recognition program capable of performing speech recognition with high accuracy using a word dictionary.

  To achieve the above object, a speech recognition device according to claim 1 is a speech recognition device that recognizes speech using a word dictionary, and is input from speech input means that receives input of a speaker's speech. General-purpose dictionary for storing a general-purpose dictionary that is a dictionary indicating correspondence between a plurality of words and information related to pronunciation of the plurality of words for each of a plurality of categories First recognition means for recognizing the voice based on the voice information acquired by the voice information acquisition means using the general-purpose dictionary stored in the storage means; and words belonging to the plurality of categories Classifying a plurality of words belonging to a reference category that is a category having a number equal to or less than a predetermined amount and relating the pronunciation of the plurality of words and the plurality of words related to each classification from the general dictionary The voice information acquired by the voice information acquisition unit using the division dictionary stored in the division dictionary storage unit that stores a plurality of division dictionaries created by extracting a part of the correspondence with the information A second recognition means for recognizing the voice based on the information; a first recognition result as a recognition result of the first recognition means; and a second recognition result as a recognition result of the second recognition means. And a result determining means for determining the speech recognition result.

  In the speech recognition apparatus of the invention according to claim 2, in addition to the configuration of the invention of claim 1, the result determination means includes the first recognition result obtained by the first recognition means and the second recognition. When the second recognition result by the means is obtained, the second recognition result including the same word as the word belonging to the reference category included in the first recognition result is determined as the recognition result. It is characterized by.

  In the speech recognition device of the invention according to claim 3, in addition to the configuration of the invention according to claim 2, the result determining means selects a word that is the same as the word belonging to the reference category included in the first recognition result. When there is no second recognition result including, the first recognition result and the second recognition result are determined as the recognition result.

  In the speech recognition device of the invention according to claim 4, in addition to the configuration of the invention according to any one of claims 1 to 3, the result determination means obtains the second recognition result by the second recognition means. If not, the first recognition result by the first recognition means is determined as the recognition result.

  In the speech recognition apparatus of the invention according to claim 5, in addition to the configuration of the invention according to any one of claims 1 to 4, the result determining means obtains the first recognition result by the first recognition means. If not, the second recognition result by the second recognition means is determined as the recognition result.

  In the speech recognition device of the invention according to claim 6, in addition to the configuration of the invention according to any one of claims 1 to 5, the plurality of divided dictionaries respectively include different words belonging to the reference category. And

  A speech recognition apparatus according to a seventh aspect of the invention includes the same word as the word belonging to the reference category included in the first recognition result, in addition to the configuration of the invention according to any one of the second to sixth aspects. In the case where there is no second recognition result, there is further provided a re-speech instruction means for outputting information for instructing the speaker to speak again to an information output means for outputting information for the speaker.

  In addition to the configuration of the invention according to any one of claims 1 to 6, the speech recognition device according to an eighth aspect of the present invention provides an information output unit that outputs a predetermined question to the speaker. Question output means for outputting is further provided, wherein the reference category is a category related to a response to the predetermined question presented by the question presentation means.

  In addition to the configuration of the invention described in claim 7, the speech recognition apparatus of the invention according to claim 9 further includes question presentation means for causing the information output means to output a predetermined question for the speaker, and the reference category Is a category related to a response to the predetermined question presented by the question presenting means.

  According to a tenth aspect of the present invention, in addition to the configuration of the eighth aspect of the invention, the speech recognition apparatus further includes a notification destination storage unit that stores a notification destination corresponding to each of the plurality of words belonging to the reference category. Referring to, the notification destination specifying means for specifying the notification destination corresponding to the word belonging to the reference category included in the recognition result determined by the recognition result determination means, and the specification specified by the notification destination specifying means A speech recognition apparatus, further comprising notification means for performing notification processing based on the recognition result for a notification destination.

  According to an eleventh aspect of the present invention, a speech recognition program causes a computer to function as various processing means of the speech recognition apparatus according to any one of the first to tenth aspects.

  According to the speech recognition apparatus of the first aspect of the present invention, a speech recognition result (first recognition result) using a general dictionary for a plurality of words and a divided dictionary in which a part of the general dictionary is extracted and created The recognition result is determined based on the used speech recognition result (second recognition result). Although the first recognition result using the general-purpose dictionary cannot be said to have high reliability as a whole, it is considered that the reliability of the words in the reference category in which the number of belonging words is equal to or less than a predetermined amount is high to some extent. On the other hand, since the divided dictionary is obtained by extracting only a part of the general-purpose dictionary according to the result of further classifying the words belonging to the reference category, the number of included words is narrowed down. Therefore, if the utterance content corresponds to a word in the divided dictionary, the second recognition result using the divided dictionary is considered to be highly reliable as a whole. Therefore, by determining the speech recognition result of the speaker based on both the first recognition result using the general-purpose dictionary and the second recognition result using the divided dictionary, a highly accurate recognition result is finally obtained. Can be obtained.

  According to the speech recognition apparatus of the second aspect of the present invention, it is considered that the reliability of the words in the reference category in which the number of belonging words is equal to or less than a predetermined amount in the first recognition result using the general-purpose dictionary is somewhat high. On the other hand, the second recognition result using the divided dictionary in which the number of words is reduced is considered to be higher in overall reliability than the first recognition result if the utterance content corresponds to the words in the divided dictionary. It is done. Therefore, in addition to the effect of the invention described in claim 1, by selecting a second recognition result including the same word as a reference based on a word having high reliability in the first recognition result, it is more reliable as a whole. A highly recognizable recognition result can be obtained.

  According to the speech recognition apparatus of the invention of claim 3, in the first recognition result using the general-purpose dictionary, it is considered that the reliability of the words in the reference category in which the number of belonging words is equal to or less than a predetermined amount is high to some extent. On the other hand, the second recognition result using the divided dictionary in which the number of words is narrowed is considered to have higher reliability as a whole than the first recognition result. Therefore, in addition to the effect of the invention described in claim 2, when the recognition results of the words of the reference category are different from each other, it is highly possible that a part of the recognition results is a correct recognition result by adopting both.

  According to the speech recognition device of the invention according to claim 4, when the second recognition result is not obtained, that is, when the recognition with the divided dictionary with the reduced number of words fails, the general dictionary is used. The first recognition result is determined as the recognition result. Therefore, in addition to the effect of the invention according to any one of claims 1 to 3, it is possible to reduce the possibility that recognition is impossible even when the utterance content does not correspond to the divided dictionary.

  According to the speech recognition device of the fifth aspect of the present invention, when the first recognition result is not obtained, the second recognition result using the divided dictionary is determined as the recognition result. Therefore, in addition to the effect of the invention according to any one of claims 1 to 4, the possibility of being unrecognizable can be reduced.

  According to the speech recognition device of the invention of claim 6, the plurality of divided dictionaries are created so as to include different words belonging to the reference category, respectively. Therefore, in addition to the effect of the invention according to any one of claims 1 to 5, since the number of words included in each divided dictionary can be minimized, the recognition accuracy can be further improved. In addition, since the divided dictionaries are created so as to include different words, recognition results specific to the words can be obtained.

According to the speech recognition device of the invention according to claim 7, when there is no second recognition result including the same word as the word belonging to the reference category included in the first recognition result, the speaker is uttered again. Information for instructing to output is output to the information output means. Therefore, in addition to the effect of the invention according to any one of claims 2 to 6, there is a high possibility that a correct recognition result can be obtained instead of a recognition result whose reliability is unknown.

  In the speech recognition device according to the eighth aspect, the reference category is a category related to a response to a predetermined question for the speaker. When a predetermined question is presented to the speaker, the response to the question typically includes words that belong to a fixed category. Therefore, in addition to the effect of the invention according to any one of claims 1 to 6, the recognition accuracy of the response by the speaker can be increased.

  In the speech recognition device according to the ninth aspect, the reference category is a category related to a response to a predetermined question for the speaker. When a predetermined question is presented to the speaker, the response to the question typically includes words that belong to a fixed category. Therefore, in addition to the effect of the invention according to the seventh aspect, the recognition accuracy of the response by the speaker can be improved.

  According to the speech recognition device of the invention of claim 10, the corresponding notification destination is specified based on the words belonging to the reference category included in the recognition result, and information based on the recognition result is transmitted. Therefore, in addition to the effect of the invention described in claim 8 or 9, the person who has received the notification can know information about the speaker without directly interacting with the speaker.

  The speech recognition program of the invention according to claim 11 can cause a computer to function as various processing means of the speech recognition apparatus according to any one of claims 1 to 10. Therefore, the effect of the invention according to any one of claims 1 to 10 can be achieved.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, embodiments of the invention will be described with reference to the drawings. The drawings to be referred to are used for explaining the technical features that can be adopted by the present invention, and the configuration of the apparatus and the flowcharts of various processes described are not intended to be limited thereto. This is just an illustrative example.

  First, referring to FIG. 1 to FIG. 8, as an example of a speech recognition system according to the present embodiment, the entire configuration of the reception system 1, and the reception device 10 as a speech recognition device that is a component of the reception system 1 and The configuration of the user terminal 20 will be described in order.

  First, an outline of the overall configuration of the reception system 1 will be described with reference to FIG. FIG. 1 is a system configuration diagram showing a schematic configuration of the reception system 1. The reception system 1 is a system that is installed in, for example, a building or a company and performs reception work for visitors. In the present embodiment, the reception system 1 is described as being installed in the company 5.

  As shown in FIG. 1, the reception system 1 includes a reception device 10 and a plurality of user terminals 20, which are connected to each other by a LAN 9. The receiving device 10 and the user terminal 20 may be general-purpose computers such as personal computers, or may be dedicated devices. In the present embodiment, description will be made assuming that the device is a dedicated device. Further, the LAN 9 may be another network, whether wired or wireless.

  Here, a procedure of processing performed in the entire reception system 1 of the present embodiment will be briefly described. When a visitor to the company 5 approaches the reception device 10 provided near the entrance, it is detected by a human sensor 109 (see FIG. 2) provided in the reception device 10. Here, since the question regarding the name of the visitor and the person in charge of the visitor (hereinafter simply referred to as the person in charge) is issued from the reception device 10, the visitor responds to the reception device 10. Then, the visitor's response is recognized by voice, and the person in charge is specified based on the recognition result. And the notification which notifies that the visitor arrived is transmitted from the reception apparatus 10 with respect to the user terminal 20 which a person in charge uses. In this way, the reception device 10 can automatically perform reception work in the company 5. The process briefly described so far will be described in detail later.

  Next, the configuration of the reception device 10 will be described with reference to FIGS. FIG. 2 is a block diagram showing an electrical configuration of the accepting device 10. FIG. 3 is an explanatory diagram of a storage area provided in the hard disk device 15. FIG. 4 is an explanatory diagram of the language model 1521. FIG. 5 is an explanatory diagram of the general-purpose dictionary 1531. FIG. 6 is an explanatory diagram of the division dictionary 1541. FIG. 7 is an explanatory diagram of the employee database 1551.

  First, the electrical configuration of the accepting device 10 will be described with reference to FIG. As shown in FIG. 2, the receiving apparatus 10 includes a CPU 101 and a control circuit unit 100 including a ROM 102 and a RAM 103 connected to the CPU 101. An input / output (I / O) interface 104 is connected to the CPU 101. A display 106, a microphone 107, a speaker 108, a human sensor 109, a communication device 110, and a hard disk device (HDD) 150 are connected to the I / O interface 104.

  The CPU 101 controls the entire receiving device 10. The ROM 102 stores programs necessary for basic operation of the receiving device 10 and setting values for the programs. The CPU 101 controls the operation of the receiving apparatus 10 according to programs stored in the ROM 102 and the HDD 150. The RAM 103 is a storage device for temporarily storing various data.

  The display 106 is a display device including a liquid crystal panel and a drive circuit. On the display 106, for example, an image of an assumed person or character who is engaged in voice guidance work, a sentence corresponding to voice uttered from the speaker 108, or the like is displayed. The microphone 107 is a device to which sound is input, and converts the input sound into sound data and outputs the sound data. The speaker 108 is a device that converts input sound data into sound and outputs the sound. In the present embodiment, the microphone 107 corresponds to “audio input means” of the present invention, and the speaker 108 and the display 106 correspond to “information output means”.

  The human sensor 109 is a sensor that detects a human body in a predetermined area in front of the reception device 10, that is, a visitor. As the human sensor 109, for example, a well-known human sensor such as an infrared sensor that emits infrared rays to a human body and detects the human body based on a change in the amount of received infrared light can be employed. The communication device 110 is a device that transmits and receives data to and from an external device such as the user terminal 20 via the LAN 9.

  The HDD 150 will be described with reference to FIGS. As shown in FIG. 3, the HDD 150 as a storage device is provided with a plurality of storage areas. The plurality of storage areas include, for example, an acoustic model storage area 151, a language model storage area 152, a general dictionary storage area 153, a divided dictionary storage area 154, an employee database (DB) storage area 155, and a program storage area 156. In this embodiment, the general dictionary storage area 153 corresponds to “general dictionary storage means” of the present invention, the divided dictionary storage area 154 corresponds to “divided dictionary storage means”, and the employee DB storage area 155 Corresponds to “notification destination storage means”.

  The acoustic model storage area 151 stores a known acoustic model (not shown) used in the speech recognition process. Although detailed description is omitted, the acoustic model is a statistical model of the acoustic characteristics (for example, frequency characteristics) of speech. For example, for each vowel and consonant, the phoneme corresponding to the feature amount is used. It is expressed with.

  The language model storage area 152 stores a language model 1521 (see FIG. 4) used in the speech recognition process. The language model defines restrictions on word connections, that is, connections between words. As typical language models, there are, for example, a description grammar model that describes the connection between words in grammar (grammar), and a statistical model (for example, word N-gram) that defines the connection between words with probability.

  Among these typical language models, the description grammar model is defined by manually describing an assumed sentence pattern in advance as an acceptable sentence pattern. Therefore, although the number of acceptable sentence patterns to be described is limited, if the utterance corresponds to the defined sentence pattern, high-accuracy speech recognition is possible. In this embodiment, since speech recognition is performed for an utterance in a relatively limited situation such as an utterance of a person who has visited the company 5, a description grammar model is adopted as a language model. However, the language model is not necessarily a descriptive grammar model, and a statistical model may be used.

  In the language model storage area 152 (see FIG. 3), patterns of acceptable sentences created in advance according to various situations assumed in the dialogue between the receiving device 10 and the visitor are stored as language models. Yes. For example, the language model 1521 shown in FIG. 4 is an example of a sentence pattern that can be accepted as a response to a question regarding the names of visitors and persons in charge. A sentence pattern can be defined, for example, by sequentially arranging categories to which words constituting a sentence belong.

  In the example of FIG. 4, the words belonging to the categories “company name”, “connection”, “visitor name”, “end 1”, “person in charge”, “honorific name”, and “end 2” are sequentially connected. And words belonging to the categories "unnecessary word", "company name", "connection", "visitor name", "end 1", "person in charge", "honorific title", and "end 2" A plurality of acceptable sentence patterns are defined, including two kinds of sentences connected in order.

  Note that FIG. 4 illustrates only the language model 1521 corresponding to the response to the question regarding the names of the visitor and the person in charge, but the language model storage area 152 also includes, for example, questions related to the business. A plurality of language models created in advance corresponding to various scenes such as a scene of a response to may be stored.

  The general dictionary storage area 153 stores a general dictionary 1531 (see FIG. 5) used for speech recognition. As shown in FIG. 5, the general-purpose dictionary 1531 describes words belonging to a category and information related to pronunciation of each word for each of a plurality of categories related to a response to a question issued by the receiving device 10. In FIG. 5, katakana is used as information related to pronunciation in order to simplify the notation, but actually, it is stored as a phoneme string in the general-purpose dictionary. For example, when the Roman alphabet is used for the phoneme string, the pronunciation corresponding to the word “that” is described as “Ano” in katakana in FIG. 5, but is actually stored as “ano”.

  FIG. 5 shows general categories including “unnecessary words”, “company name”, “connection”, “visitor name”, “end 1”, “person in charge”, “honorific name”, and “end 2” as categories. An example of a dictionary is shown. That is, for a category included in the sentence defined in the language model 1521 described above, this is a dictionary in which words belonging to each category and their pronunciation are described. For the category in the general dictionary 1531, any of the words belonging to the category (hereinafter referred to as a non-important category) that does not affect the meaning of the whole sentence even if any word belonging to the category is used. There are categories where the meaning of the whole sentence changes (hereinafter referred to as important categories) depending on whether or not is used.

  In the example of the general dictionary 1531 in FIG. 5, the categories “unnecessary words”, “connection”, “end 1”, “title”, and “end 2” are non-important categories, and the categories “company name”, “visitor” “Name” and “Name of person in charge” are important categories.

  In the general-purpose dictionary 1531, the number of words belonging to the important category “company name” is 5,000, the number of words belonging to “visitor name” is 10,000, and the number of words belonging to “person in charge” is 15. That is, the number of words of “company name” and “visitor name” is significantly larger than the category “person in charge”. The number of persons in charge of company 5 is limited, but for one person in charge, for example, there are a number of business partners who come to the meeting, and there are also a number of persons who belong to these companies. There may be. Therefore, in the general dictionary 1531, in order to cover all the words, the number of words of “company name” and “visitor name” becomes very large in this way. In FIG. 5, the number of words belonging to each category is shown for illustrative purposes only, and the actual general-purpose dictionary 1531 may not include the number of words.

  As described above, in addition to the language model 1521 (see FIG. 4), when there are other language models created in advance according to various situations, the general-purpose dictionary stored in the general-purpose dictionary storage area 153 is This includes information on words and pronunciations for each category corresponding to other language models.

  In the divided dictionary storage area 154 (see FIG. 3), a plurality of divided dictionaries (see FIG. 6) used for speech recognition are stored. The divided dictionary classifies words belonging to a category having a predetermined number of words or less among the important categories included in the general-purpose dictionary 1531 described above. From the general-purpose dictionary 1531, words of other categories related to each classification are classified. It can be created by extracting. In the following description, among the important categories included in the general-purpose dictionary 1531, a category that is used as a classification criterion and has a predetermined number of words or less is referred to as a “reference category”.

  The reference category may be a category having the number of words belonging to a predetermined number or less, or a category having a ratio with respect to the number of words belonging to another category being a predetermined value or less. It is more preferable to use an important category having the smallest number of words. Moreover, it is preferable that the word to which it belongs is a category relevant to the word of another category, respectively. Furthermore, it is desirable that the category is included in the response to the question presented to the visitor from the reception device 10. In this case, since the number of words included in each divided dictionary can be minimized, the recognition accuracy can be further improved. In addition, since the divided dictionaries are created so as to include different words, recognition results specific to the words can be obtained.

  In the present embodiment, among the important categories “company name”, “visitor name”, and “person in charge” included in the general dictionary 1531, “person in charge” has the smallest number of words. Therefore, the category “name of person in charge” of the general dictionary 1531 is set as the reference category. The words (name of the person in charge) belonging to this reference category are classified for each person in charge, and belong to the categories “company name” and “visitor name” according to the company and visitor to which each person in charge relates. Words are assigned to each category. A divided dictionary is created by associating words belonging to each category with information related to their pronunciation, and stored in the divided dictionary storage area 154.

  For example, FIG. 6 shows an example of the division dictionary 1541 created for the classification in which the person in charge of the company 5 is “Kato”. The person in charge “Kato” is in charge of two companies with the names “A Kogyo” and “B Transportation” and two visitors with the names “Nagamitsu” and “Go”. A word is selected from the general dictionary 1531 and included in the divided dictionary 1541 as a word belonging to the categories “company name” and “visitor name”. Note that the words in the non-important category are related to any classification, and are included in the divided dictionary 1541 as they are from the general-purpose dictionary 1531.

  As a result, as shown in FIG. 6, the number of words of the important categories “company name”, “visitor name”, and “person in charge” are 2, 2, and 1, respectively. That is, in the divided dictionary 1541, the number of words of “company name” and “visitor name” can be drastically reduced as compared with the general dictionary 1531. In FIG. 6, the number of words belonging to each category is shown for illustrative purposes only, and the actual divided dictionary 1541 may not include the number of words.

  In this embodiment, since there are 10 words belonging to the category “person in charge” in the general-purpose dictionary 1531 (see FIG. 5), when a divided dictionary based on the person in charge is created as described above, The number of divided dictionaries is 10 including 6 divided dictionaries 1541. However, even when a division dictionary is created based on the person-in-charge name, the number of words in the person-in-charge name included in each division dictionary need not necessarily be one. For example, it is possible to classify each person in charge who belongs to the same department in the company 5 or each person in charge who is engaged in the same business, and create a divided dictionary including one to several names. In this case, the number of divided dictionaries may be less than 10.

  The employee DB storage area 155 (see FIG. 3) stores an employee DB 1551 (see FIG. 7) that stores personal information of all employees of the company 5 (hereinafter referred to as employee information). The employee information is created for each employee, and includes, for example, a name, last name, and contact information as shown in FIG. As the contact information, for example, the IP address of the user terminal 20 used by the employee, the email address of the employee, a telephone number, or the like can be employed. The employee information may include information such as an employee code for identifying each employee and a department to which the employee belongs, in addition to the information shown in FIG. In the present embodiment, the name or last name stored in the employee DB 1551 corresponds to the “word” of the present invention, and the contact address corresponds to the “notification destination” of the present invention.

  In the program storage area 156 (see FIG. 3), for example, a program for controlling various operations of the receiving apparatus 10 including a program used for various processes described later, a set value, and the like are stored. For example, a program stored in a CD-ROM is installed via a CD-ROM drive (not shown) and stored in the program storage area 156. Alternatively, a program downloaded from outside connected to the LAN 9 or another network (not shown) via the communication device 110 may be stored. Although not shown, the HDD 150 also stores display data and audio data that are transmitted to the user terminal 20 in a process that will be described later.

  Next, the configuration of the user terminal 20 will be described with reference to FIG. FIG. 8 is a block diagram showing an electrical configuration of the user terminal 20. As illustrated in FIG. 8, the user terminal 20 includes a CPU 201 and a control circuit unit 200 including a ROM 202 and a RAM 203 connected to the CPU 201. An input / output (I / O) interface 204 is connected to the CPU 201. A display 206, a microphone 207, a speaker 208, a communication device 210, and a hard disk device (HDD) 250 are connected to the I / O interface 204. That is, the configuration of the user terminal 20 is the same as that of the reception device 10 except that there is no human sensor.

  The CPU 201 controls the entire user terminal 20. The ROM 202 stores programs necessary for basic operations of the user terminal 20 and setting values for the programs. The CPU 201 controls the operation of the user terminal 20 according to a program stored in the ROM 202 or the HDD 250. The RAM 203 is a storage device for temporarily storing various data.

  The display 206 is a display device including a liquid crystal panel and a drive circuit. The display 206 displays, for example, a notification sentence that announces the arrival of the visitor. The microphone 207 is a device to which sound is input, and converts the input sound into sound data and outputs the sound data. The speaker 208 is a device that converts input sound data into sound and outputs the sound. For example, the speaker 208 outputs a notification sound that notifies the arrival of the visitor. The HDD 250 stores a program for controlling various operations of the user terminal 20, setting values, and the like.

  Hereinafter, various processes performed in the reception apparatus 10 configured as described above will be described with reference to FIGS. 9 to 17. FIG. 9 is a flowchart of the main process of the receiving apparatus 10. FIG. 10 is a flowchart of the agency process executed in the main process. FIG. 11 is a flowchart of the voice recognition process. 12 and 13 are flowcharts of the recognition result determination process executed in the speech recognition process. 14 to 17 are explanatory diagrams illustrating specific examples of recognition results obtained by voice recognition. The processing shown in FIGS. 9 to 13 is executed by the CPU 101 of the accepting apparatus 10 according to the program stored in the program storage area 156 of the HDD 150.

  The main process shown in FIG. 9 is started when the power of the receiving apparatus 10 is turned on, and is continuously performed until the power is turned off. When the process is started, first, both a start flag and an end flag are set to false and stored in a flag storage area (not shown) of the RAM 103 (S1). The start flag is a flag that defines a period for recognizing a visitor's utterance input from the microphone 107 of the receiving apparatus 10 in a voice recognition process (see FIG. 11) described later. Specifically, the utterance is recognized while the start flag is set to true, and the recognition is stopped when the start flag is set to false. The end flag is a flag indicating whether or not to end the speech recognition process. Specifically, the speech recognition process is continued while the end flag is false, and ends when the end flag is true.

  Subsequently, a voice recognition process is activated (S2). Specifically, when the voice recognition program stored in the program storage area 156 of the HDD 150 is activated, the voice recognition process (see FIG. 11) is started and executed in parallel with the main process. . The voice recognition process will be described later in detail.

  It is determined whether or not a visitor is detected by the human sensor 109 (S3). While a visitor is not detected, it will be in a standby state until a visitor is detected (S3: NO). When a visitor is detected by the human sensor 109 (S3: YES), an intermediary process is performed (S4 and FIG. 10). Although details will be described later, in the intermediary process, the person in charge who responds to the visitor is specified based on the recognition result obtained by the voice recognition process (see FIG. 11) that is executed separately, and the user used by the person in charge A notification process for the terminal 20 is performed.

  If the power is not turned off after the agency process (S5: NO), the process returns to step S3, and the agency process is repeated for each visitor. Then, when the power supply of the receiving device 10 is turned off (S5: YES), the end flag stored in the flag storage area of the RAM 103 is set to false in order to end the voice recognition process being executed in parallel. It is changed to true (S6), and the main process shown in FIG. 9 ends.

  With reference to FIG. 10, the agency process performed in the main process will be described. As shown in FIG. 10, when the mediation process is started, first, the re-utterance flag is set to false and stored in the flag storage area of the RAM 103 (S41). The recurrence flag is a flag indicating whether or not the visitor has already been prompted to recite the same content when the visitor's utterance recognition fails. Specifically, when the recurrent talk flag is true, it indicates that the visitor has already been prompted to reoccurrence, and when false, it indicates that the recurrent talk has not been prompted yet.

  Subsequently, the CPU 101 causes the speaker 108 to output a sound of content asking questions about the visitor name and the person in charge (S42). More specifically, the text data of the question text that is created in advance and stored in a predetermined storage area (not shown) of the HDD 150 is read out. Then, the sound data is converted into sound data, and the sound data is converted into sound by the speaker 108 and output. Here, for example, a voice requesting the visitor name and the name of the person in charge is output, such as "Please give me the name of the customer and the name of the person in charge."

  In order to recognize the utterance when the visitor responds to this question, the start flag set to false at the beginning of the main process is changed to true (S43). Thereby, in the speech recognition processing (see FIG. 11) executed in parallel, the recognition of the utterance is started.

  A final recognition result (hereinafter referred to as a final result) obtained in the speech recognition process is stored in a predetermined storage area of the RAM 103. Therefore, the CPU 101 obtains the final result stored in the RAM 103 and determines whether or not it indicates a recognition failure (S45). If the final result indicating the recognition failure is stored in the RAM 103 (S45: YES), it means that the response to the above-mentioned question could not be recognized. Therefore, the CPU 101 determines whether or not the recurrence flag stored in the flag storage area of the RAM 103 is false (S46).

  When the re-utterance flag is false (S46: NO), the visitor only responds to the question once, and never makes the same response again. Therefore, the sound of the content prompting the re-utterance is output from the speaker 108 (S47). Also here, text data of a text created in advance and stored in a predetermined storage area (not shown) of the HDD 150 is read and converted into voice data, and further converted into voice by the speaker 208 and output. Here, for example, a voice requesting the name of the visitor and the name of the person in charge is output, such as “Please ask again”. Then, after the recurrence flag is changed to true indicating that the visitor has already been prompted to recite the same content (S48), the process returns to step S43.

  In step S43 following step S48, in order to recognize the recurrent speech by the visitor, the start flag set to false after recognition of the first response in the voice recognition processing (see FIG. 11) being performed in parallel, It is changed to true (S43). Thereby, in the voice recognition process, recognition of the recurrent utterance of the visitor is started and the final result is acquired (S44).

  If the final result of the recurrent speech also indicates a recognition failure (S45: YES), the recurrent speech flag is set to true in step S48 of the previous process (S46: NO). Therefore, the CPU 101 determines that the name of the person in charge of this visitor is the name of the representative person in charge without performing any further processing for asking the same question to the visitor, and sets the name of the representative person in charge in a predetermined storage area of the RAM 103. Store (S49). Here, the representative person-in-charge name is a name of a person who is predetermined to respond to a visitor when the person in charge is unknown. For example, the representative name may be stored in the employee DB 1551 so that the representative person name among the person names can be identified. Alternatively, it may be stored in a predetermined storage area of the HDD 150 separately.

  On the other hand, when the first response of the visitor to the question of the visitor and the person in charge is successfully recognized (S45: NO), or when the second response is recognized after being prompted to recurrence again (S45). : NO), the final result includes the name of the person in charge. In other words, the person in charge can be notified of the arrival of the visitor. Therefore, the CPU 101 reads the text data of the text that has been created in advance and stored in a predetermined storage area (not shown) of the HDD 150, and causes the speaker 108 to output a voice requesting standby (S51). Here, for example, a voice message “Please wait for a while because the person in charge has been taken over” is output.

  Subsequently, the CPU 101 specifies the person in charge name based on the final result (S52). Specifically, the person-in-charge name included in the final result stored in the RAM 103 is extracted. As described above, in the present embodiment, a visitor's response is recognized using the language model 1521 (see FIG. 4), which is a descriptive grammar model. The word "person in charge" is included. Although details will be described later, a plurality of final results may be stored in the RAM 103 as a result of the speech recognition process. In such a case, the person in charge included in each final result may be specified. The identified person-in-charge name is stored in a predetermined storage area of the RAM 103.

  In step S53, the contact information of the person in charge who responds to the visitor is specified. Specifically, the CPU 101 refers to the employee DB 1551 (see FIG. 7) stored in the employee DB storage area 155 and matches the person-in-charge name specified in step S49 or S52 and stored in the RAM 103. Identify contact information that corresponds to the employee's last name. Then, notification processing for notifying the arrival of the visitor is performed for the contact information of the identified person in charge (S54). As described above, when a plurality of persons in charge are specified, the contacts of all persons in charge are specified, and the notification process may be performed on the contacts of all persons in charge.

  For example, when the IP address of the user terminal 20 used by the person in charge is specified as the notification destination, the CPU 101 transmits audio data with content notifying the arrival of the visitor to the user terminal 20, and the speaker 208 of the user terminal 20. The sound may be output from. Here, for example, it may be notified only based on a predetermined notice, for example, “You are here”, or it is included in the final result, such as “You are Ami of A Industry”. The company name and visitor name may also be notified. In addition, when an e-mail address is specified as a notification destination, text data of a notification sentence may be transmitted to the address. In this case, a notification sentence is displayed on the display 206 of the user terminal 20.

  By such notification processing, the person in charge who has received the notification can know information that the visitor has arrived or who has visited without directly interacting with the visitor. After the notification process for the contact information of the person in charge is performed, the relay process shown in FIG. 10 ends, and the process returns to the main process in FIG.

  Next, with reference to FIG. 11, the speech recognition process started in step S2 of the main process shown in FIG. 9 and executed in parallel with the main process will be described.

  As shown in FIG. 11, in the speech recognition process, first, an acoustic model, a language model, and a dictionary are each read from a predetermined storage area of the HDD 150 (S21). Specifically, an acoustic model (not shown) is read from the acoustic model storage area 151. The language model 1521 (see FIG. 4) corresponding to the response to the question regarding the visitor name and the person in charge name is read from the language model storage area 152. A general dictionary 1531 (see FIG. 5) is read from the general dictionary storage area 153, and a plurality of division dictionaries created based on the general dictionary 1531 including the division dictionary 1541 (see FIG. 6) are divided dictionary storage areas. Read from 154.

  Subsequently, it is determined whether or not the end flag stored in the flag storage area of the RAM 103 is false (S22). As described above, the end flag remains false unless the receiving apparatus 10 is turned off (S22: YES). In this case, it is determined whether or not the start flag stored in the flag storage area is true (S23).

  As described above, the start flag is set after the question voice of the visitor name and the person in charge name is output from the speaker 108 in the intermediary process of the main process executed when the visitor is detected by the human sensor 109. , True (FIG. 10, S43). That is, it remains false until the visitor's name and the person in charge of the question name are output to the visitor (S23: NO). In this case, the standby state is set until the end flag is set to true or the start flag is set to true (S22: YES, S23: NO).

  When the power of the receiving apparatus 10 is turned off and the end flag is set to true in step S6 of the main process in FIG. 9 (S22: NO), the voice recognition process shown in FIG. If the start flag is set to true in step S43 of the agency process in FIG. 10 (S23: YES). The voice of the visitor's response to the visitor name and the person in charge name questions is input from the microphone 107, and the voice data is acquired in the RAM 103 (S24).

  Voice recognition of the acquired voice data is performed (S25). That is, the speech data is converted into text using an acoustic model, a language model 1521 and a dictionary, that is, a plurality of divided dictionaries including a general dictionary 1531 and a divided dictionary 1541. Specifically, for example, after the voice data is analyzed and the feature amount is extracted, matching between the acoustic model and the language model 1521 is performed. The language model 1521 refers to a plurality of divided dictionaries including a general dictionary 1531 and a divided dictionary 1541 as dictionaries.

  As a result of matching, the likelihood is obtained for each sentence acceptable by the language model 1521, and the sentence with the highest likelihood is obtained as the recognition result. In addition, when the likelihood becomes a value equal to or less than a predetermined threshold, the recognition result cannot be obtained as a recognition failure. When speech recognition is successful, the recognition result (text) obtained can be identified, and when speech recognition fails, the information indicating the recognition failure is the result obtained by referring to which dictionary. As a result, the result is stored in the storage area of the RAM 103.

  For example, as described above, when 10 divided dictionaries are created for every 10 persons in charge, the number of dictionaries is 11 in total, and thus 11 recognitions using the language model 1521 are performed. It will be. That is, if recognition is successful in all dictionaries, 11 recognition results (text) are obtained corresponding to each dictionary. As the identification information of the result corresponding to each dictionary stored, for example, the result corresponding to the general dictionary is assigned zero (0), and the result corresponding to the divided dictionary of 10 is assigned the numbers 1 to 10, respectively. It is stored in the RAM 103 in association with (text or information indicating recognition failure). The 11 types of speech recognition performed with reference to each dictionary may be performed simultaneously in parallel or sequentially. Considering the waiting time of visitors, it is preferable to carry out in parallel.

  After the voice recognition is performed in this way (S25), a recognition result determination process for determining the final result is performed (S26 and FIG. 12). As described above, the final result determined in the recognition result determination process is used to specify the person in charge in the intermediary process (see FIG. 10) of the main process executed in parallel.

  After the recognition result determination process (S26), the start flag stored as true in the flag storage area is changed to false in order to temporarily stop the voice recognition and wait for the next utterance by the visitor (S27). . Then, the process returns to step S22. When the power supply of the receiving apparatus 10 is turned off and the end flag is set to true (FIG. 9, S5: YES, S6), the voice recognition process shown in FIG. 11 ends. While the power is not turned off and the end flag is false (S22: NO), the processes in steps S22 to S27 are repeated as described above.

  The recognition result determination process will be described with reference to FIGS. In the following, ten divided dictionaries are created for every ten persons in charge, and zero (0) for the results corresponding to the general-purpose dictionary and 1 to 10 for the results corresponding to the ten divided dictionaries. In the following description, it is assumed that an identification number is assigned.

  As shown in FIG. 12, when the process is started, it is first determined whether or not there is a recognition result obtained by referring to the general dictionary (hereinafter simply referred to as “general dictionary recognition result”) (S261). ). Specifically, it is determined whether the result stored in the RAM 103 corresponding to the identification number zero (0) is a recognition result (text) or information indicating a recognition failure. When the information indicating the recognition failure is stored, it is determined that there is no recognition result of the general dictionary (S261: NO), and the recognition result obtained by referring to the divided dictionary (hereinafter simply referred to as “divided dictionary”). It is determined whether or not there is a "recognition result" (S262). Specifically, it is determined whether or not there is at least one recognition result (text) among the results stored in the RAM 103 corresponding to the identification numbers 1 to 10.

  If at least one of the results corresponding to the identification numbers 1 to 10 is a recognition result (text), it is determined that there is a recognition result of the divided dictionary (S262: YES). In this case, the recognition result of the divided dictionary is determined as the final result (S263), the recognition result determination process in FIG. 12 is terminated, and the process returns to the voice recognition process shown in FIG. When there are a plurality of recognition results, all the recognition results are the final results. As described above, there is no recognition result of the general dictionary, but when there is a recognition result of the divided dictionary, the recognition result of the divided dictionary is used as a final result, thereby reducing the possibility of a recognition failure finally. . In addition, when there are a plurality of recognition results, a correct recognition result may be included in the final result by adopting all the recognition results as the final result.

  For example, it is assumed that the result of speech recognition of the visitor's utterance (FIG. 11, S25) and the result shown in FIG. That is, information indicating a recognition failure is stored as a result corresponding to the identification number zero (0) which is the result of the general dictionary. In addition, the results corresponding to the identification numbers 1 to 10 as the result of the divided dictionary include both information indicating recognition failure and the recognition result. For example, the result of the division dictionary corresponding to the identification number 1 is information indicating a recognition failure, but the result corresponding to the identification number 2 is the recognition result (text) “I am Emitsu of A Industry, but please Mr. Sato” It is. In addition, the result corresponding to the identification number 10 is a recognition result (text) “E-Nagaku Nakamichi, please Sakami-san”.

  In this example, there is a general dictionary recognition result (S261: YES in FIG. 12), and it is further determined that there is a divided dictionary recognition result (S262: YES). Therefore, if the recognition results of the identification numbers 3 to 9 all indicate recognition failure, the two types of recognition results corresponding to the identification numbers 2 and 10 are determined as the final results. . Then, in the intermediary process of the main process (see FIG. 10) that is executed separately, “Sato” and “Sako” are specified as the person in charge from these two types of final results, and the users who use these two people Notification is made to the terminal 20.

  On the other hand, if all the results stored in the RAM 103 corresponding to the identification numbers 1 to 10 are information indicating a recognition failure in step S262, it is determined that there is no recognition result of the divided dictionary (S262: NO). ). In this case, it means that the recognition has failed in both the speech recognition performed with reference to the general dictionary and the speech recognition performed with reference to the divided dictionary. Therefore, the final result is determined as a recognition failure (S264), the recognition result determination process in FIG. 12 ends, and the process returns to the speech recognition process shown in FIG. In this case, in the relay process of the main process (see FIG. 10), the visitor is prompted to re-speak. If the final result is a recognition failure even after recognizing the recurrent utterance, a representative name predetermined as the person in charge is specified, and a notification is sent to the user terminal 20 used by the representative. Done.

  In step S261, if the result stored in the RAM 103 corresponding to the identification number zero (0) is the recognition result (text), it is determined that there is a general dictionary recognition result (S261: YES). ). Further, it is determined whether there is a recognition result of the divided dictionary (S271). If all the results stored in the RAM 103 corresponding to the identification numbers 1 to 10 are information indicating a recognition failure, it is determined that there is no recognition result of the divided dictionary (S271: NO). In this case, the recognition result of the general dictionary is determined as the final result (S272). In this way, if the utterance content does not correspond to the divided dictionary and the recognition fails in the divided dictionary, the possibility that the final result cannot be obtained is reduced by adopting the recognition result of the general dictionary as the final result. it can.

  When it is determined that there is a recognition result of the general dictionary (S261: YES), at least one of the results corresponding to the identification numbers 1 to 10 is the recognition result, and there is also a recognition result of the divided dictionary (S271: YES). Then, it is determined whether or not the number of recognition results of the divided dictionary is greater than 1 (S273 in FIG. 13). When the number of recognition results of the divided dictionary is larger than 1 (S273: YES), the recognition result of a plurality of divided dictionaries is obtained in addition to the recognition result of the general dictionary. As described above, since the divided dictionary is a dictionary created by dividing the general-purpose dictionary for each person in charge, the number of words is narrowed down, and it is considered that the reliability of the entire recognition result is high. Therefore, processing is performed in which the recognition result with higher reliability among the recognition results of the plurality of divided dictionaries is used as the final result. Specifically, the person in charge name included in the recognition result of the general dictionary is specified as person in charge name 1 (S279). Then, among the recognition results of the plurality of divided dictionaries, the recognition result including the person in charge name 1 is determined as the final result (S280).

  The general-purpose dictionary includes all the names of persons in charge required by visitors, and the number of words in person-in-charge is smaller than the number of words in other important categories. Therefore, it is considered that the person-in-charge name included in the recognition result of the general dictionary has high reliability. On the other hand, since the division dictionary is created for each person in charge, if the person in charge of the division dictionary is included in the utterance of the visitor, the person name included in the recognition result is highly reliable. If not, the reliability is not high. Therefore, by using the recognition result of the divided dictionary including the same person name as the person name included in the general-purpose dictionary as the final result, it is possible to obtain a final result with high reliability as a whole.

  For example, it is assumed that the result of speech recognition of the visitor's utterance (S25 in FIG. 11) and the result shown in FIG. That is, as a result corresponding to the identification number zero (0), which is the result of the general dictionary, “K Kogyo Nakamichi, please Sato-san please” is stored. In addition, as a result corresponding to the identification numbers 1 to 10 which are the results of the divided dictionary, recognition results (text) are stored corresponding to at least the identification numbers 1, 2 and 10. Specifically, the result of the division dictionary corresponding to the identification number 1 is a recognition result (text) “I am Nagai of K Industry, but please Mr. Kato”. The result of the division dictionary corresponding to the identification number 2 is a recognition result (text) “I am Emitsu of A Industry, but please Sato-san”. The result of the division dictionary corresponding to the identification number 10 is a recognition result (text) “E-Nagashi's middle road, please Sakami-san”.

  In this example, there is a recognition result of the general dictionary (S261: YES in FIG. 12), and it is further determined that there are a plurality of recognition results of the divided dictionary (S271: YES, S273: YES in FIG. 13). Then, “Sato”, the name of the person in charge included in “K Kogyo Nakamichi, please Sato, please”, which is the recognition result of the general-purpose dictionary, is specified as person-in-charge name 1 (S279). In addition, the recognition result of the divided dictionary including “Sato” identified as the person in charge 1 is “A Industry's Nagamitsu, but please Mr. Sato,” corresponding to the identification number 2 as the final result. (S280). Then, in the intermediary process of the main process (see FIG. 10) that is executed separately, “Sato” is identified as the person in charge from this final result, and the user terminal 20 used by Mr. Sato is notified. .

  On the other hand, if it is determined in step S273 that the number of recognition results of the divided dictionary is 1 (S273: NO), the person-in-charge name included in the general-purpose dictionary recognition result is the person-in-charge name as in step S279. After being identified as 1 (S274), the person-in-charge name included in the recognition result of the divided dictionary is identified as person-in-charge name 2 (S275). Then, it is determined whether or not the person in charge name 1 and the person in charge name 2 match (S276). As described above, since the name of the person in charge included in the recognition result of the general dictionary is highly reliable, if the name of the person in charge included in the recognition result of the divided dictionary is the same as this, the reliability of the recognition result of this divided dictionary is confirmed. This is because the property is high. Therefore, if the person in charge name 1 and the person in charge name 2 are the same (S276: YES), the recognition result of the divided dictionary is determined as the final result (S277). Thereby, the final result with high reliability as a whole can be obtained.

  For example, it is assumed that the result of speech recognition of the visitor's utterance (FIG. 11, S25) and the result shown in FIG. That is, as a result corresponding to the identification number zero (0), which is the result of the general dictionary, “K Kogyo Nakamichi, please Sato-san please” is stored. Further, the result of the division dictionary corresponding to the identification number 2 is a recognition result (text) “I am Emitsu of A Industry, but please Mr. Sato”. It is assumed that all other divided dictionary results are recognition failures.

  In this example, there is a recognition result of the general dictionary (FIG. 12, S261: YES), and it is further determined that there is only one recognition result of the divided dictionary (S271: YES, S273: NO in FIG. 13). . Then, “Sato”, which is the name of the person in charge included in the recognition result of the general dictionary, is specified as the person in charge name 1 (S274), and “Sato”, which is the name of the person in charge included in the recognition result of the divided dictionary, is the name of the person in charge. 2 (S275). Since person-in-charge name 1 and person-in-charge name 2 are the same (S276: YES), the recognition result of the division dictionary corresponding to identification number 2 is "Emitsu of A Industry, but please Sato-san", The final result is determined (S277). Then, in the intermediary process of the main process (see FIG. 10) that is executed separately, “Sato” is identified as the person in charge from this final result, and the user terminal 20 used by Mr. Sato is notified. .

  On the other hand, when it is determined in step S276 that the person-in-charge name 1 as the person-in-charge name included in the recognition result of the general dictionary and the person-in-charge name 2 as the person-in-charge name included in the recognition result of the divided dictionary are different. (S276: NO), both the recognition result of the general dictionary and the recognition result of the divided dictionary are determined as final results (S278). As mentioned above, this is because the person name included in the recognition result of the general dictionary is highly reliable, so if the person name included in the recognition result of the divided dictionary is different from this, the recognition result of the divided dictionary is better. However, the reliability is not necessarily higher than the recognition result of the general-purpose dictionary. Therefore, by adopting both as the final result, it is possible to cope with whichever is correct.

  For example, it is assumed that the result of speech recognition of the visitor's utterance (FIG. 11, S25) and the result shown in FIG. That is, as a result corresponding to the identification number zero (0) which is the result of the general dictionary, “K Kogyo Nakamichi, please Kato-san please” is stored. Further, the result of the division dictionary corresponding to the identification number 2 is a recognition result (text) “I am Emitsu of A Industry, but please Mr. Sato”. It is assumed that all other divided dictionary results are recognition failures.

  In this example, there is a recognition result of the general dictionary (FIG. 12, S261: YES), and it is further determined that there is only one recognition result of the divided dictionary (S271: YES, S273: NO in FIG. 13). . Then, “Kato” that is the name of the person in charge included in the recognition result of the general dictionary is specified as the person in charge name 1 (S274), and “Sato” that is the name of the person in charge included in the recognition result of the divided dictionary is 2 (S275). Since the person in charge name 1 and the person in charge name 2 do not match (S276: NO), the division result corresponding to the identification number 2 is “K Kogyo Nakamichi, please Kato-san,” which is the recognition result of the general-purpose dictionary. The dictionary recognition result, "I am Emitsu of A Industry, but please Sato-san", is determined as the final result (S277). Then, in the intermediary process of the main process (see FIG. 10) that is separately executed, “Kato” and “Sato” are specified as the person in charge from these two types of final results, and the user terminals used by these two persons 20 is notified.

  As described above, in the reception system 1 according to the present embodiment, when a visitor is detected by the reception device 10, a voice for asking a visitor name and a person in charge name is output. When the visitor responds to the question, the voice is recognized by referring to the general dictionary and the divided dictionary. The final recognition result is determined based on the recognition result when speech recognition is performed with reference to the general dictionary and the recognition result when speech recognition is performed with reference to the divided dictionary. The name of the person in charge included in the final recognition result is specified, and the arrival of the visitor is notified to the contact information.

  Although the recognition result of the general-purpose dictionary cannot be said to be highly reliable as a whole, it is considered that the person-in-charge name, which is a word in the reference category having the number of words belonging to a predetermined amount or less, has a certain degree of reliability. On the other hand, since the divided dictionary is obtained by extracting only a part of the general-purpose dictionary for each person in charge, the number of included words is narrowed down. Therefore, if the utterance content corresponds to the name of the person in charge included in the divided dictionary, the recognition result of the divided dictionary is considered to be highly reliable as a whole. Therefore, as in the present embodiment, the recognition result of the visitor's utterance is determined based on both the recognition result of the general dictionary and the recognition result of the divided dictionary, thereby finally obtaining a highly accurate recognition result. Can do.

  In the present embodiment, the CPU 101 that acquires the voice input from the microphone 107 in step S25 of FIG. 11 corresponds to the “voice information acquisition unit” of the present invention. The CPU 101 that performs speech recognition using the general-purpose dictionary and the divided dictionary in step S25 corresponds to “first recognition means” and “second recognition means”. The CPU 101 that determines the final result in steps S263, S264, and S272 in FIG. 12 and steps S277, S278, and S280 in FIG. 13 corresponds to “recognition result determining means”.

  In step S47 in FIG. 10, the CPU 101 that causes the speaker 108 to output a voice prompting the visitor to respond again is equivalent to “recurrent speech instruction means”. The CPU 1010 that causes the speaker 108 to output a question regarding the visitor name and the person in charge in step S42 in FIG. 10 corresponds to “question presenting means”. The CPU 101 that specifies the notification destination of the person in charge in step S53 corresponds to “notification destination specifying means”, and the CPU 101 that performs notification processing in step S54 corresponds to “notification means”.

  It should be noted that the configuration and processing shown in the above-described embodiment are examples, and it goes without saying that various modifications are possible. For example, in the recognition result determination process (FIGS. 12 and 13) according to the above-described embodiment, the recognition result of the general dictionary and the recognition result of the divided dictionary are obtained, but the person-in-charge names included in the recognition results do not match. (S276 in FIG. 13: NO) As the final results, the names of the two persons in charge included in each are adopted as the final results. However, in this case, the name of the person in charge is considered to be highly reliable in the recognition result of the general dictionary, but the person in charge of the recognition result of the divided dictionary is different from that. , And voice recognition may be redone.

  Specifically, the final result may be determined as a recognition failure in step S278 in FIG. As a result, in the intermediary process shown in FIG. 10, assuming that the final result is a recognition failure (S45: YES), a recurrent talk is prompted (S47). This increases the possibility of obtaining a correct final result rather than a final result whose reliability is unknown.

  In the recognition result determination process, when the recognition result of the general-purpose dictionary and the recognition results of a plurality of divided dictionaries are obtained (S261: YES, S271: YES in FIG. 12, S273: YES in FIG. 13), the final result The recognition result of the divided dictionary including the person-in-charge name that matches the person-in-charge name 1 included in the recognition result of the general-purpose dictionary is adopted (S279 to S280). When there are a plurality of recognition results of the divided dictionary, it is usually considered that one of the person-in-charge names included therein matches the person-in-charge name 1 included in the recognition result of the general dictionary. However, if the person-in-charge names included in the recognition results of the plurality of division dictionaries are all different from the person-in-charge name 1, the recognition results of the general dictionary and the recognition results of the plurality of division dictionaries are the same as in step S278 of FIG. All may be the final result, or the visitor may be encouraged to speak again.

  Further, in the recognition result determination process (FIGS. 12 and 13), when a plurality of final results are determined, a plurality of person names are specified, and a notification process is performed for the notification destinations of all the persons in charge. Has been done. However, it is not always necessary to notify all of the plurality of specified persons in charge, and when a plurality of final results are obtained, for example, it is possible to notify a predetermined representative person in charge. . Alternatively, if there is a visitor reservation database that stores visitor reservations in association with the person in charge, the visitor included in the final result in the reservation database among the specified names of the persons in charge The name of the person in charge stored corresponding to the name may be specified, and the person in charge may be notified.

  In the above-described embodiment, each of the plurality of divided dictionaries including the general-purpose dictionary 1531 (see FIG. 5) and the divided dictionary 1541 (see FIG. 6) includes all the words of the non-important category. However, the words belonging to the non-important category do not necessarily need to be included in the general-purpose dictionary or the divided dictionary, and another dictionary including only the words of the non-important category is created and used together with the general-purpose dictionary or the divided dictionary for speech recognition. You may make it refer.

  In the above-described embodiment, the reception apparatus 10 includes the microphone 107, the speaker 108, and the human sensor 109, and all processes such as visitor detection, voice input / output, and voice recognition are performed by the same apparatus. However, a part of the configuration of the receiving apparatus 10 according to the above-described embodiment may be another apparatus. For example, even if an interphone type reception terminal that includes a microphone, a speaker, and a human sensor and is connected to the reception device 10 is installed near the entrance of the company 5, the reception device 10 may be installed at another location in the company 5. Good. Then, the reception terminal may detect a visitor or input a voice, and the information may be transmitted to the reception device 10. Moreover, what is necessary is just to transmit audio | voice data from the reception apparatus 10 to a reception terminal, and to output a sound from the speaker of a reception terminal. In this case, the reception apparatus 10 does not need to include the microphone 107, the speaker 108, and the human sensor 109.

  In the above-described embodiment, the receiving device 10 is provided with the HDD 150, but information (general dictionary 1531, division dictionary 1541, employee DB 1551, etc.) stored in the HDD 150 is received via the LAN 9, for example. 10 may be stored in a separate storage device that can be connected to the computer 10 and necessary information may be read out during processing.

  In the above-described embodiment, the question or recurrent speech instruction presented to the visitor from the reception device 10 is performed by outputting sound from the speaker 108. However, it is possible to display a question or an instruction for a recurrent story on the display 106. Also, audio output and display may be performed simultaneously.

  The above embodiment is an example in which the reception device 10 performs voice recognition of a visitor in the company or building reception system 1 and notifies the user terminal 20 used by the person in charge according to the recognition result. The invention can also be applied to other devices. For example, the present invention may be applied to a guidance device that is installed at a railway station, recognizes a user's utterance by voice recognition, and provides facility guidance along the line.

In this case, the guidance device may be, for example, “Which station is nearby?” Or “Which facility information”
It is assumed that the user utters a response including the station name and facility name. Therefore, the general dictionary needs to cover these words included in the response to the question. Since there are an enormous number of facilities along the entire route, recognition may be inaccurate if user speech recognition is performed using only a general dictionary. On the other hand, the number of station names on a certain line is limited to at most about several tens, which is significantly smaller than the number of facility names. Therefore, it is possible to classify the station name using the station name as a reference category and extract words belonging to other categories related to the station name (for example, facility names, transfer routes, etc.) for each station name classification to create a divided dictionary. . Thereby, the number of words, such as a facility name contained in each division | segmentation dictionary, can be reduced significantly.

  And like the above-mentioned embodiment, using a general-purpose dictionary including all station names, facility names, etc., and a divided dictionary created using station names as reference categories, a predetermined name such that the station name is included in the response from the guidance device. May be issued, and voice recognition of the user's response to it may be performed. Thereby, compared with the case where speech recognition is performed using only a general-purpose dictionary, a highly accurate recognition result can be obtained.

  Furthermore, instead of the employee DB 1551 (see FIG. 7) of the above-described embodiment, a notification destination database for storing a notification destination corresponding to each facility in association with the facility name can be prepared in advance. Then, based on the facility name included in the recognition result, notification processing can be performed for the corresponding notification destination. For example, the guidance device can be connected to each facility via a telephone line or other network, and connected to the facility's representative phone or facility computer based on the contact address corresponding to the facility name specified from the recognition result. Processing can be performed.

1 is a system configuration diagram illustrating a schematic configuration of a reception system 1. FIG. 3 is a block diagram showing an electrical configuration of the receiving device 10. FIG. 3 is an explanatory diagram of a storage area provided in the hard disk device 15. FIG. FIG. 10 is an explanatory diagram of a language model 1521. It is explanatory drawing of the general purpose dictionary 1531. FIG. It is explanatory drawing of the division | segmentation dictionary 1541. FIG. It is explanatory drawing of the employee database 1551. FIG. 3 is a block diagram showing an electrical configuration of a user terminal 20. FIG. 4 is a flowchart of main processing of the receiving device 10. It is a flowchart of the agency process performed by the main process. It is a flowchart of a speech recognition process. It is a flowchart of the recognition result determination process performed by voice recognition processing. FIG. 13 is a flowchart of a recognition result determination process executed in the voice recognition process, which is a continuation of FIG. 12. It is explanatory drawing which shows the specific example of the recognition result obtained by voice recognition. It is explanatory drawing which shows the specific example of the recognition result obtained by voice recognition. It is explanatory drawing which shows the specific example of the recognition result obtained by voice recognition. It is explanatory drawing which shows the specific example of the recognition result obtained by voice recognition.

Explanation of symbols

10 receiving apparatus 101 CPU
106 Display 107 Microphone 108 Speaker 150 HDD
153 General dictionary storage area 154 Divided dictionary storage area 155 Employee database storage area 20 User terminal

Claims (11)

A speech recognition device for recognizing speech using a word dictionary,
Voice information acquisition means for acquiring voice information of the voice input from voice input means for receiving input of a speaker's voice;
For each of a plurality of categories, using the general dictionary stored in a general dictionary storage means for storing a general dictionary that is a dictionary indicating correspondence between a plurality of words and information related to pronunciation of the plurality of words, the audio information First recognition means for recognizing the voice based on the voice information acquired by the acquisition means;
Of the plurality of categories, classify a plurality of words belonging to a reference category that is a category having a predetermined number of words or less, and from the general dictionary, the plurality of words related to each classification and the plurality of words Using the divided dictionary stored in the divided dictionary storage means for storing a plurality of divided dictionaries created by extracting a part of the correspondence with pronunciation information, the voice information acquisition means acquires the Second recognition means for recognizing the voice based on voice information;
A result determining means for determining a recognition result of the voice based on a first recognition result that is a recognition result of the first recognition means and a second recognition result that is a recognition result of the second recognition means; A speech recognition apparatus comprising the voice recognition device.   The result determination means is included in the first recognition result when the first recognition result by the first recognition means and the second recognition result by the second recognition means are obtained. The speech recognition apparatus according to claim 1, wherein the second recognition result including the same word as a word belonging to a reference category is determined as the recognition result.   When there is no second recognition result including the same word as the word belonging to the reference category included in the first recognition result, the result determination means includes the first recognition result and the second recognition result. The speech recognition apparatus according to claim 2, wherein the recognition result is determined as the recognition result.   The result determination means determines the first recognition result by the first recognition means as the recognition result when the second recognition result by the second recognition means is not obtained. The speech recognition apparatus according to claim 1.   The result determination means determines the second recognition result by the second recognition means as the recognition result when the first recognition result by the first recognition means is not obtained. The speech recognition apparatus according to claim 1.   The speech recognition apparatus according to claim 1, wherein the plurality of divided dictionaries include different words belonging to the reference category.   If there is no second recognition result including the same word as the word belonging to the reference category included in the first recognition result, information for instructing the speaker to speak again is given to the speaker. The speech recognition apparatus according to any one of claims 2 to 6, further comprising a recurrence instruction means for outputting information to an information output means for outputting information. A question presenting means for outputting a predetermined question for the speaker to an information output means for outputting information for the speaker;
The speech recognition apparatus according to claim 1, wherein the reference category is a category related to a response to the predetermined question presented by the question presenting unit. A question presenting means for causing the information output means to output a predetermined question for the speaker;
The speech recognition apparatus according to claim 7, wherein the reference category is a category related to a response to the predetermined question presented by the question presenting unit. With reference to a notification destination storage unit that stores a notification destination corresponding to each of the plurality of words belonging to the reference category, a word belonging to the reference category included in the recognition result determined by the recognition result determination unit A notification destination specifying means for specifying the corresponding notification destination;
10. The voice according to claim 8, further comprising a notification unit that performs a notification process based on the recognition result for the notification destination specified by the notification destination specifying unit. Recognition device.   A speech recognition program for causing a computer to function as various processing means of the speech recognition apparatus according to claim 1.

Images