JP5252119B2 - Elevator voice call registration device - Google Patents

Description

  The present invention relates to an elevator voice call registration device for registering elevator calls by voice input.

  As a voice call registration device for registering elevator calls by voice input, a device including a plurality of voice recognition processing units that perform voice recognition using different recognition methods has been proposed. This voice call registration device increases the accuracy of voice recognition by determining the recognition results of a plurality of voice recognition processing units together, and changes the response to the user according to the accuracy of voice recognition. Thereby, a comfortable use environment can be given to a user (for example, refer to patent documents 1).

  However, when the one described in Patent Document 1 is used, voice recognition is performed by the same method for all users. For this reason, speech that does not match any of the recognition methods prepared in the speech recognition processing unit cannot be recognized. A user who makes such voice cannot register an elevator call such as a destination call by voice input indefinitely. Moreover, the thing of patent document 1 cannot register an elevator call unless it is simultaneously recognized by a plurality of voice recognition processing units. That is, the one described in Patent Document 1 is unusable because the threshold for determining registration of an elevator call is uniformly raised for all users.

  On the other hand, a voice call registration device has been proposed that can accept only the voice of a user registered in advance. According to this voice call registration device, voice misrecognition can be reduced (see, for example, Patent Document 2). Furthermore, an apparatus for accumulating speech analysis results and recognizing speaker attributes based on the accumulation results has been proposed. According to these apparatuses, it is possible to improve speaker attribute identification accuracy and further reduce speech misrecognition (see, for example, Patent Documents 3 and 4).

  However, when the devices described in Patent Documents 2 to 4 are applied to a voice call registration device, it is necessary to limit the user to a specific speaker. For this reason, the utilization range of the thing of patent documents 2 thru | or 4 will become narrow. On the other hand, the elevator voice call registration device needs to recognize the voices of an unspecified number of users. For this reason, it is difficult to apply the devices described in Patent Documents 2 to 4 to the voice call registration device.

  On the other hand, devices with multiple speech recognition dictionaries (recognized words and acoustic models) are proposed according to the usage attribute, environmental attributes expressed by environmental noise, and individual attributes expressed by the gender and age of the speaker. ing. According to this device, it is possible to select an appropriate speech recognition dictionary by giving actual environment attributes and actual user attributes, and select an acoustic model that matches the characteristics of the speaker without limiting the speaker. (For example, refer to Patent Document 5).

  However, when the one described in Patent Document 5 is applied to a voice call registration device, there is an annoyance that an actual environment attribute or an actual user attribute must be given each time it is used.

  On the other hand, a display that displays information according to the use situation by determining the user attributes and the number of adult men, adult women, children, etc. based on information input from the camera device provided in the elevator A device has been proposed. If this display device discriminating method is applied to the selection of an acoustic model, it is possible to easily select an acoustic model that may suit the user without limiting the user (see, for example, Patent Document 6). .

Japanese Patent No. 3082618 Japanese Patent No. 2557939 Japanese Unexamined Patent Publication No. 10-240287 Japanese Special Table 2003-524805 Japanese Patent Laid-Open No. 2002-229584 Japanese Unexamined Patent Publication No. 2007-261722

  However, the discrimination method described in Patent Document 6 is not always accurate. For this reason, there is a problem that an acoustic model that does not match the user's characteristics may be selected.

  The present invention has been made to solve the above-described problems, and its purpose is not to limit the user, but to register the elevator call by voice input in an easy manner. It is an object of the present invention to provide an elevator voice call registration device capable of increasing the possibility of selecting an acoustic model suitable for a feature.

An elevator voice call registration device according to the present invention includes an audio input unit that captures input audio to an audio input device provided in an elevator car or a landing, and an acoustic model storage that stores a plurality of acoustic models having different acoustic characteristics. And a user information extracting unit for extracting user information relating to characteristics of a user at a car or a hall provided with the voice input device, and a position of a car or a hall provided with the voice input device An elevator information extraction unit that extracts elevator information related to the state of the elevator, and an acoustic model used when recognizing the elevator call from the input voice based on the user information and the elevator information, an acoustic model selection unit that selects from among acoustic models, every time the user information extraction unit extracts user information, extracted Comprising a user information storage unit to use user information in association with the elevator information storing user tendency information about trends in features of the user, wherein the acoustic model selection unit, based on the user tendency information The acoustic model to be used when recognizing the call of the elevator when the elevator is in a state corresponding to the elevator information is selected .

  According to the present invention, it is possible to increase the possibility of selecting an acoustic model that matches the user's characteristics when registering an elevator call by voice input in an easy manner without limiting the user.

BRIEF DESCRIPTION OF THE DRAWINGS It is a block diagram of the elevator voice call registration apparatus and elevator control part in Embodiment 1 of this invention. It is a flowchart for demonstrating operation | movement in case the elevator call registration apparatus in Embodiment 1 of this invention registers the call of an elevator. It is a flowchart for demonstrating operation | movement when the voice call registration apparatus of the elevator in Embodiment 1 of this invention selects an acoustic model. It is a block diagram of the voice call registration apparatus and elevator control part in Embodiment 2 of this invention. It is a block diagram of the voice call registration apparatus and elevator control part of the elevator in Embodiment 3 of this invention.

  A mode for carrying out the invention will be described with reference to the accompanying drawings. In addition, in each figure, the same code | symbol is attached | subjected to the part which is the same or it corresponds, The duplication description is simplified or abbreviate | omitted suitably.

Embodiment 1 FIG.
Generally, an elevator is provided in a building such as a building. A call registration device is provided in the elevator car and at the landing. Using this call registration device, the user registers an elevator call. In response to the call registered by the call registration device, the elevator car moves up and down.

  Here, as one of call registration devices, a voice call registration device that registers elevator calls by voice input has been proposed. This voice call registration device recognizes a voice uttered by a user in a car or a hall and registers an elevator call. If this voice call registration device is used, a call of an elevator can be easily registered even by a user who has both hands closed.

  In such a voice call registration device, if the voice uttered by the user is erroneously recognized, the elevator call desired by the user cannot be registered. Therefore, in the voice call registration device of the present embodiment, the voice recognition accuracy is improved. Hereinafter, the voice call registration device of the present embodiment will be described in detail.

FIG. 1 is a configuration diagram of an elevator voice call registration device and an elevator control unit according to Embodiment 1 of the present invention.
The voice call registration device of FIG. 1 is provided in an elevator car or a landing. This voice call registration device includes a voice input unit 1, an A / D conversion unit 2, a voice cutout unit 3, an acoustic analysis unit 4, an information input unit 5, a user information extraction unit 6, a building information storage unit 7, and an acoustic model selection. Unit 8, a recognition dictionary 9, a plurality of acoustic models 10, and a speech recognition unit 11.

  The voice input unit 1 has a function of taking an input voice to a voice input device (not shown) such as a microphone provided in an elevator car or a landing. The A / D conversion unit 2 has a function of converting the input voice captured by the voice input unit 1 into digital data. The voice extraction unit 3 has a function of detecting a silent section of digital data passed from the A / D conversion unit 2. In addition, the voice extraction unit 3 has a function of cutting out a section sandwiched between silent sections of digital data as speech sections to be subjected to voice recognition based on the silent section information. The acoustic analysis unit 4 has a function of converting the utterance section extracted by the speech extraction unit 3 into feature amount data used for speech recognition by arithmetic processing such as Fourier transform.

  The information input unit 5 has a function of acquiring information related to the characteristics of the user detected by a device provided in a normal elevator such as a sensor device, a scale device, or a camera device. The user information extraction unit 6 has a function of extracting user information related to the user's characteristics such as the user's back appearance and the riding speed on the car from the information input to the information input unit 5. The building information storage unit 7 has a function of storing building information. This building information includes building information relating to the characteristics of each floor of the building, such as information relating to tenants occupying each floor of a building such as a building provided with an elevator. That is, the building information storage unit 7 functions as a building information storage unit that stores building information.

  The acoustic model selection unit 8 has a function of extracting user tendency information related to a tendency of user characteristics associated with building information. This user tendency information may be stored in the acoustic model selection unit 8 or may be stored in another storage unit such as the building information storage unit 7. The acoustic model selection unit 8 has a function of estimating user attributes based on user information extracted by the user information extraction unit 6, user tendency information associated with building information, and the like.

  Various types of user attributes can be set in consideration of the detection contents and detection accuracy of user characteristics such as sensor devices, scale devices, and camera devices. For example, the user attribute may be set so as to distinguish an adult from a child. Further, the user attribute may be set so as to distinguish each age, such as teenagers and twenties. Further, the user attribute may be set so as to distinguish gender such as male or female. In addition, the user attributes may be set so as to be distinguished by complex conditions such as age and sex. The acoustic model selection unit 8 can appropriately set the weighting of each piece of information such as user information and building information when estimating user attributes.

  The recognition dictionary 9 has a function of storing words to be recognized as speech, such as “Ikkai” and “Main floor”. The plurality of acoustic models 10 are stored in each acoustic model storage device (not shown). These acoustic models 10 are composed of acoustic data in which feature amount data of each phoneme is arranged. These acoustic data correspond to user attributes estimated by the acoustic model selection unit 8 such as age and sex, and have different acoustic characteristics. These acoustic models 10 are pre-attached with tags indicating which user attribute the model corresponds to.

  The speech recognition unit 11 has a function of detecting a tag corresponding to the user attribute estimated by the acoustic model selection unit 8 from the plurality of acoustic models 10. The voice recognition unit 11 has a function of selecting the acoustic model 10 having the detected tag as the acoustic model 10 having the feature closest to the user attribute estimated by the acoustic model selection unit 8. In other words, the speech recognition unit 11 substantially extracts the acoustic model 10 selected by the acoustic model selection unit 8 from the plurality of acoustic models 10. Furthermore, the speech recognition unit 11 has a function of recognizing words described in the recognition dictionary 9 from speech input to the speech input unit 1 using the selected acoustic model 10.

  FIG. 1 also shows an elevator control unit 12. The elevator control unit 12 includes a call registration unit 13. The call registration unit 13 has a function of storing in advance the words in the recognition dictionary 9 and the floor of a building or the like provided with an elevator in association with each other. For example, “Ikkai” in the recognition dictionary 9 is literally stored with “first floor” associated therewith. “First floor” is also stored in association with “main floor” of the recognition dictionary 9. When the call registration unit 13 recognizes the floor associated with the word recognized by the voice recognition unit 11, the call registration unit 13 registers the destination call corresponding to the floor as an elevator call.

  The elevator control unit 12 also includes an elevator information management unit 14. The elevator information management unit 14 has a function of detecting and managing various elevator states. For example, the elevator information management unit 14 has a function of managing elevator information such as the position of the car or landing, the traveling direction of the car, and the door open / close state provided with the voice input device. In particular, the elevator information management unit 14 reliably detects and manages the current position (current floor) of the car that changes every moment.

  In the present embodiment, the acoustic model selection unit 8 also functions as an elevator information extraction unit that extracts elevator information from the elevator information management unit 14. And the acoustic model selection part 8 extracts the user tendency information regarding the tendency of the feature of the user matched with each elevator information. This user tendency information is grasped by investigating elevator users for a certain period. This user tendency information may be stored in the acoustic model selection unit 8 or may be stored in another storage unit.

  The acoustic model selection unit 8 recognizes the user tendency information extracted from the building information as the first user tendency information, and recognizes the user tendency information extracted from the elevator information as the second user tendency information. The acoustic model selection unit 8 estimates the user attribute in consideration of not only the user information and the first user tendency information but also the second user tendency information.

Next, using FIG. 2 and FIG. 3, the operation when the voice call registration device of the present embodiment registers an elevator call will be described.
FIG. 2 is a flowchart for explaining the operation when the elevator voice call registration device according to the first embodiment of the present invention registers an elevator call. FIG. 3 is a flowchart for explaining the operation in the case where the elevator voice call registration device according to Embodiment 1 of the present invention selects an acoustic model.

First, an outline of a procedure for registering an elevator call will be described with reference to FIG.
That is, in step S1, when voice is input into the car or landing voice input device, the voice input unit 1 captures the voice and proceeds to step S2. In step S2, the A / D converter 2 converts the sound into digital data, and the process proceeds to step S3.

  In step S3, the voice cutout unit 3 cuts out and detects the speech section of the digital data, and proceeds to step S4. In step S4, the acoustic analysis unit 4 converts the speech section into feature amount data used for speech recognition, performs acoustic analysis, and proceeds to step S5. In step S5, the acoustic model selection unit 8 substantially selects the acoustic model 10 and proceeds to step S6.

  In step S <b> 6, the speech recognition unit 11 compares the feature amount data acoustically analyzed by the acoustic analysis unit 4 with the acoustic data of the acoustic model 10 selected by the acoustic model selection unit 8. Thereby, the input voice taken into the voice input unit 1 is recognized, and the process proceeds to step S7. In step S7, the voice recognition unit 11 outputs the recognition result of the input voice to the call registration unit 13 of the elevator control unit 12, and the process proceeds to step S8.

  In step S8, the call registration unit 13 determines whether or not the floor associated with the word related to the recognition result has been recognized. If the floor associated with the word related to the recognition result is not recognized, the operation ends. That is, the destination call is not registered. On the other hand, if the floor associated with the word related to the recognition result is recognized, the process proceeds to step S9. In step S9, the call registration unit 13 registers a destination call corresponding to the floor, and the operation ends.

  Here, in step S5 of FIG. 3, as described above, not only the user information and the first user tendency information but also the second user tendency information is considered, and the call of the elevator is recognized from the input voice. The acoustic model 10 to be used is selected. Hereinafter, the selection procedure of the acoustic model 10 in the present embodiment will be specifically described with reference to FIG.

  First, in step S11, the acoustic model selection unit 8 refers to the building information in the building information storage unit 7 and extracts the first user tendency information associated with the building information. Then, it progresses to step S12 and the acoustic model selection part 8 refers to the elevator information of the elevator information management part 14, and extracts the 2nd user tendency information matched with the present elevator information.

  Then, it progresses to step S13 and the user information extraction part 6 extracts user information. Thereafter, the acoustic model selection unit 8 refers to the user information and proceeds to step S14. In step S14, the acoustic model selection unit 8 estimates the user attribute in consideration of not only the user information and the first user tendency information but also the second user tendency information, and the process proceeds to step S15. In step S15, the acoustic model selection unit 8 substantially selects the acoustic model 10 that matches the user attribute, and the operation ends.

  According to the first embodiment described above, not only user information and building information but also elevator information is considered, and the acoustic model 10 used when recognizing an elevator call is selected. For this reason, it is possible to increase the possibility of selecting the acoustic model 10 that matches the characteristics of the user when the elevator call is registered by voice input in an easy manner without limiting the user.

  The selection of the acoustic model 10 can increase the recognition accuracy for the voice uttered by the user, and can make it difficult to react to the useless talk other than the talker having the same characteristics as the user and the voice flowing from the announcement device. . That is, it is possible to increase the possibility of preventing misrecognition of speech uttered by the user and call misregistration due to the misrecognition.

  The user information extraction unit 6 extracts user information from user characteristics detected by devices such as a sensor device, a scale device, and a camera device that are generally provided in an elevator. Therefore, it is possible to increase the possibility of selecting the acoustic model 10 that matches the user's characteristics by an easy method without adding a special device.

  In the first embodiment, the selected acoustic model 10 is used, speech recognition is performed only once, and elevator calls are registered. However, a threshold value for the recognition likelihood may be provided, and the word and likelihood recognized as a recognition result of the speech recognition unit 11 may be output to determine whether or not to register the elevator call. And when the recognition likelihood in the acoustic model 10 selected as a thing suitable for a user is low, the change of the acoustic model 10 used for speech recognition may be repeated until the recognition likelihood exceeds a threshold value. .

  In the first embodiment, the recognition result of the input voice using the acoustic model 10 selected based on the user information, the building information, and the elevator information is output, and the elevator call is registered. It was. However, among the recognition results of the input speech using all the acoustic models 10, the recognition result of the input speech using the acoustic model 10 selected based on the user information, the building information, and the elevator information is output. You may make it register the call of an elevator.

Embodiment 2. FIG.
FIG. 4 is a configuration diagram of an elevator voice call registration device and an elevator control unit according to Embodiment 2 of the present invention. In addition, the same code | symbol is attached | subjected to the part which is the same as that of Embodiment 1, or an equivalent, and description is abbreviate | omitted.

  The voice call registration device according to the second embodiment is obtained by adding a user information storage unit 15 to the voice call registration device according to the first embodiment. The user information storage unit 15 stores the extracted user information in association with the elevator information each time the user information extraction unit 6 extracts the user information, thereby storing the second user tendency information. It has a function to accumulate.

  That is, in the second embodiment, the user information storage unit 15 learns the second user tendency information in conjunction with the elevator information. The learning result is reflected in the selection of the acoustic model 10. Specifically, the acoustic model selection unit 8 considers not only the user information and the first user tendency, but also the second user tendency information automatically accumulated during the actual operation of the elevator, and thus the elevator. It is set to select the acoustic model 10 to be used when recognizing the call.

  And the acoustic model selection part 8 of this Embodiment increases the weight of 2nd user tendency information with the increase in the accumulation | storage amount of 2nd user tendency information, and the weight of 1st user tendency information. Is set to decrease. For example, the acoustic model selection unit 8 is set to decrease the weight of the first user tendency information in inverse proportion to the learning amount of the second user tendency information.

  In the voice call registration device having such a configuration, the voice recognition unit 11 substantially selects the acoustic model 10 to be used when recognizing the elevator call whenever the elevator information changes. Switch to acoustic model 10. Then, the voice recognition unit 11 recognizes the input voice captured by the voice input unit 1 using the acoustic model 10 that is switched each time the elevator information changes.

  According to the second embodiment described above, the second user tendency information corresponding to the elevator information accumulated during the actual operation of the elevator is also taken into consideration, and the elevator is in a state corresponding to each elevator information. The acoustic model 10 to be used is selected. For this reason, the possibility of selecting the acoustic model 10 that matches the characteristics of the user can be further increased.

  Here, if the user's tendency changes with the change of the tenant etc. who moved into the building, if the building information is not updated, the acoustic model 10 suitable for the user may be selected at the beginning of the change of the tenant etc. Go down. However, in the second embodiment, as the accumulated amount of the second user tendency information corresponding to each elevator information increases, the weight of the second user tendency information increases and the first corresponding to the building information. The weight of user tendency information is reduced. For this reason, if the accumulated amount of the second user tendency increases by continuing the operation of the elevator, there is a possibility of selecting the acoustic model 10 suitable for the user even if the user forgets to manually update the building information. Can be high.

Embodiment 3 FIG.
FIG. 4 is a configuration diagram of an elevator voice call registration device and an elevator control unit according to Embodiment 3 of the present invention. In addition, the same code | symbol is attached | subjected to the part which is the same as that of Embodiment 1 or 2, or an equivalent, and description is abbreviate | omitted.
In the voice call registration device of the third embodiment, instead of the information input unit 5, the user information extraction unit 6 and the building information storage unit 7 of the first embodiment, a voice feature extraction unit 16, a voice feature storage unit 17, An input speech learning unit 18 is provided.

  The speech feature extraction unit 16 has a function of extracting speech features from the feature amount data converted by the acoustic analysis unit 4. The characteristics of this voice are classified into adult voice, child voice, female voice, male voice and the like. That is, in the third embodiment, the voice feature extraction unit 16 functions as a user information extraction unit that extracts user information from the features of the user's input voice.

  The voice feature storage unit 17 stores the second user tendency information by storing the voice feature of the user in association with the elevator information each time the voice feature is extracted by the voice feature extraction unit 16. Is provided. That is, in the third embodiment, the voice feature storage unit 17 functions as an equivalent to the user information storage unit of the second embodiment.

  In the input speech learning unit 18, an unspecified speaker model is set as an initial setting of the acoustic model 10 used when the elevator is in a state corresponding to each elevator information. Then, the input speech learning unit 18 learns what kind of speech features the user has in each elevator state based on the second user tendency information accumulated in the speech feature storage unit 17. . That is, the input speech learning unit 18 gradually changes the setting of the acoustic model 10 used when the elevator is in a state corresponding to each elevator information by the learning.

  And the acoustic model selection part 8 of Embodiment 3 uses the acoustic model used when recognizing an elevator call based on the second user tendency information corresponding to each elevator information learned by the input speech learning part 18. 10 is selected to be selected from a plurality of acoustic models 10.

  In the voice call registration device having such a configuration, the voice recognition unit 11 substantially selects the acoustic model 10 to be used when recognizing the elevator call whenever the elevator information changes. Switch to acoustic model 10. Then, the voice recognition unit 11 recognizes the input voice captured by the voice input unit 1 using the acoustic model 10 that is switched each time the elevator information changes.

  According to the third embodiment described above, even if the first user tendency information associated with the user information and the building information from the sensor device, the scale device, and the camera device is not extracted, the user characteristics It is possible to increase the possibility of selecting the acoustic model 10 that suits the above.

  As described above, according to the elevator voice call registration device of the present invention, it is suitable for the characteristics of the user when registering the elevator call by voice input in an easy manner without limiting the user. It can be used for an elevator that increases the possibility of selecting an acoustic model.

DESCRIPTION OF SYMBOLS 1 Voice input part 2 A / D conversion part 3 Voice extraction part 4 Acoustic analysis part 5 Information input part 6 User information extraction part 7 Building information storage part 8 Acoustic model selection part 9 Recognition dictionary 10 Acoustic model 11 Voice recognition part 12 Elevator Control unit 13 Call registration unit 14 Elevator information management unit 15 User information storage unit 16 Speech feature extraction unit 17 Speech feature storage unit 18 Input speech learning unit

Claims (7)

A voice input unit that captures input voice to a voice input device provided in an elevator car or a landing;
An acoustic model storage unit storing a plurality of acoustic models having different acoustic characteristics;
A user information extraction unit for extracting user information relating to the characteristics of the user at the car or the landing provided with the voice input device;
An elevator information extraction unit for extracting elevator information related to the state of the elevator including the position of a car or a landing provided with the voice input device;
Based on the user information and the elevator information, an acoustic model selection unit that selects an acoustic model to be used when recognizing the elevator call from the input voice from the plurality of acoustic models;
A user information storage unit that stores user tendency information relating to a tendency of user characteristics by associating the extracted user information with the elevator information each time the user information extraction unit extracts user information; ,
Equipped with a,
The acoustic model selection unit selects, based on the user tendency information, an acoustic model to be used when recognizing the elevator call when the elevator is in a state corresponding to the elevator information. Voice call registration device for elevators. A speech recognition unit that outputs a recognition result of the input speech using the acoustic model selected by the acoustic model selection unit;
The elevator voice call registration device according to claim 1, further comprising: A speech recognition unit that outputs the recognition result of the input speech using the acoustic model selected by the acoustic model selection unit among the recognition results of the input speech using the plurality of acoustic models;
The elevator voice call registration device according to claim 1, further comprising:   The user information extraction unit extracts the user information from user characteristics detected by at least one of a sensor device, a scale device, and a camera device provided in the elevator. The elevator voice call registration device according to any one of claims 3 to 4. A building information storage unit that stores building information relating to the characteristics of each floor of the building in which the elevator is provided;
With
The acoustic model selection unit, on the basis of the user tends information and the building information, claims 1 to 4, characterized in that selecting an acoustic model to be used in recognizing the call of the elevator The elevator voice call registration device according to any one of the above. The acoustic model selection unit increases the weighting of the user tendency information and decreases the weighting of the building information when selecting an acoustic model as the accumulated amount of the user tendency information increases. The elevator voice call registration device according to claim 5 . The user information extraction unit, an elevator voice call registration device according to any one of claims 1 to 6, characterized in that extracting the user information from the characteristics of the input speech.

Images