JP2005140860A - Speech recognizing device and its control method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To guide a microphone for silence speech recognition in such a manner that the microphone is fitted to an optimum position. <P>SOLUTION: The speech input is formed from a microphone worn by a user (S 501) and the acoustic featured values relating to the input speech are extracted (S 502). Next, an acoustic likelihood is calculated by using a previously created acoustic model for identifying the wearing position (S 503). The information relating to the propriety of the wearing position of the microphone is presented (S 504) by using the calculated acoustic likelihood. A user is able to judge whether the microphone is mounted in the appropriate position based on the presentation. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a voice recognition device and a control method thereof, and in particular, a voice for performing voice recognition using a microphone configured to collect a non-audible voice of a user by wearing the voice recognition device at a predetermined part of the user. The present invention relates to a recognition device and a control method thereof.

  A speech recognition technique called non-speech recognition has been proposed (for example, Non-Patent Document 1). This is not intended for normal speech handled by conventional speech recognition, but for speech that cannot be heard by third parties (whispering), whispering like a monologue without vocal cord vibration. Speech recognition technology. In this voiceless recognition, a special surface-bonding microphone such as a stethoscope is attached near the back of the ear. The non-audible sound uttered by the speaker is transmitted to the body surface through the body and collected by a special microphone. Since the sound signal collected in this way is not easily affected by surrounding environmental sounds, good recognition performance can be expected even in an actual environment. In addition, since silence is hardly audible to surrounding humans, it is excellent in secrecy.

Nakajima Yuki et al., "Speechless recognition by extracting conducted sound in weak body", Acoustical Society of Japan 2003 Spring Conference Presentation Proceedings I, p175-176, March 2003)

  As described above, a special microphone used for speechless recognition is attached near the back of the speaker's ear, but the position where non-audible speech can be successfully collected is limited. In other words, the microphone mounting position in silent recognition is a very important factor in recognizing non-audible speech, and the voice recognition accuracy largely depends on the microphone mounting position. However, the user does not always wear the microphone at an appropriate position, and a case where the user wears the microphone at a shifted position is also conceivable. Considering such a case, it can be said that a mechanism for guiding the microphone to be mounted at an optimum position is necessary. Alternatively, it is necessary to introduce a technique for preventing the recognition rate from greatly decreasing even when the microphone is mounted at a position shifted from an appropriate position.

  According to one aspect of the present invention, for example, voice recognition is performed using a microphone configured to collect weak voices that are not accompanied by vocal fold vibration of the user by being worn on a predetermined part of the user. A voice recognition device, an acoustic model created based on voice collected from a speaker in a state where the microphone is mounted at an optimal sound collection position, and voice input by a user wearing the microphone, Calculation means for calculating an acoustic likelihood based on the acoustic model, and a presentation means for presenting information regarding the suitability of the current mounting position of the microphone to the user using the acoustic likelihood calculated by the calculation means. There is provided a voice recognition device characterized by comprising:

  According to the present invention, it is possible to guide the microphone for silent recognition to be mounted at an appropriate position.

  DESCRIPTION OF EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram illustrating a configuration of a speech recognition apparatus according to the present embodiment.

  101 is a central processing unit (CPU) that controls the entire apparatus, 102 is a control memory (ROM), and 103 is a random access memory (RAM). Reference numeral 104 denotes an external storage device such as a hard disk device or a nonvolatile memory, 105 denotes an information input unit such as a button or a keyboard, 106 denotes an information output unit that displays the state of the device or provides voice guidance, 107 denotes a voice input unit, and 108 denotes This is a bus that connects the above-described units. Specifically, the voice input unit 107 is a microphone configured to collect a non-audible voice of the user by wearing the voice input unit 107 at a predetermined part of the user (for example, near the back of the ear). The speech input unit 107 for such non-audible speech includes a paper “Non-speech recognition by extraction of weak body conduction sound” by Nakajima et al. , Pp. 175-176) and JP 2000-57325 A can be used. FIG. 14 shows a configuration example of the sound collection unit of the voice input unit 107. For example, the sound collecting unit is configured to record the vibration of the diaphragm 1 with the condenser microphone 2. This diaphragm 1 is used by adhering it to the speaker's body surface (for example, behind the ear and in the vicinity of the neck). Even in the case of a silent voice, the vibration is transmitted from the body to the body surface, and thus it is possible to pick up the silent voice.

  The ROM 102 stores a voice recognition program as shown, and the external storage device 104 stores various acoustic models described later. However, the voice recognition program may be installed in the external storage device 104 instead of the ROM 102.

  In this apparatus, the CPU 101 executes a speech recognition program stored in the ROM 102, thereby performing speech recognition processing on the speech data input from the speech input unit 107 and displaying a recognition result on the information output unit 105.

  The configuration of the speech recognition apparatus in the present embodiment is generally as described above, but each process by the speech recognition program will be described below.

  FIG. 2 is a diagram showing a module configuration of a speech recognition program involved in the acoustic model creation process in the present embodiment.

  In this acoustic model creation process, two types of acoustic models are used: an acoustic model used for speech recognition (recognition acoustic model) and an acoustic model used to identify the mounting position of the microphone (mounting position identification acoustic model). Is created. The feature amount calculation processing module 201 extracts an acoustic feature amount from the input voice data. The data storage module 202 records data in the external storage device 104. The mounting position identification acoustic model creation module 203 creates a mounting position identification acoustic model using the feature quantity calculated by the feature quantity calculation processing module 201. The recognition acoustic model creation module 204 creates a speech recognition acoustic model using the feature amount calculated by the feature amount calculation processing module 201. Here, in the mounting position identification acoustic model creation module 203 and the recognition acoustic model creation module 204, both acoustic models may be created by the same method using the same audio data. A compact acoustic model that calculates acoustic likelihood for a more limited speech is suitable for the model. On the other hand, although the acoustic model for recognition differs depending on the recognition performance specification, it is preferable that the acoustic model is acoustically more detailed than the acoustic model for identifying the mounting position.

  FIG. 3 is a flowchart showing a flow of acoustic model creation processing in the present embodiment.

  First, in step S301, the voice of an unspecified speaker is input. Here, it is assumed that the microphone is mounted at a position where the non-audible sound can be collected most effectively (hereinafter referred to as “optimum sound collection position”). Next, in step S <b> 302, the acoustic feature quantity is extracted from the voice data by the feature quantity calculation processing module 201 and stored in the external storage device 104 by the data storage module 202. In step S303, the mounting position identification acoustic model and the recognition acoustic model are obtained by the mounting position identification acoustic model creation module 203 and the recognition acoustic model creation module 204, respectively, using the feature amount calculated in step S302. These two acoustic models are created and stored in the external storage device 104 by the data storage unit 202.

  Through the above processing, voice is collected from an unspecified speaker with the microphone mounted at the optimum sound collection position, and based on this, two acoustic models are created: a mounting position identification acoustic model and a recognition acoustic model. It was.

  Next, the microphone mounting position confirmation process in the present embodiment will be described. This process is executed when a microphone is attached, and this process is intended to allow the user to adjust the position of the microphone when the position of the microphone is inappropriate.

  FIG. 4 is a diagram showing a module configuration of a speech recognition program involved in the microphone mounting position confirmation process in the present embodiment.

  The audio input module 401 receives input of audio data from the attached microphone. The feature quantity calculation processing module 402 extracts an acoustic feature quantity from the input voice data. The likelihood calculation processing module 403 performs likelihood calculation using the extracted feature quantity and the mounting position identification acoustic model. The acoustic model storage module 404 stores the acoustic model in the external storage device 104. The status display module 405 causes the information output unit 106 to display drawing, sound, and the like corresponding to the calculated acoustic likelihood. As a result, the user can determine whether or not the mounting position of the microphone is appropriate. The information input processing module 406 inputs feedback that the user performs to the system through buttons, a keyboard, and the like.

  FIG. 5 is a flowchart showing the flow of the microphone attachment position confirmation process in the present embodiment.

  First, in step S501, the voice input module 401 performs voice input from a microphone attached to the user. The input voice to be uttered at this time differs depending on the acoustic model for identifying the mounting position to be used.In the case of an acoustic model created under the same conditions as the acoustic model for recognition, it is not necessary to be the specified utterance, In the case of the created acoustic model, only limited utterances are accepted as input speech. In addition, it is necessary to always input the same voice as the input voice while identifying the mounting position. In step S501, when there is no voice input, the state is maintained while the input state is maintained, and when there is a voice input, the process proceeds to the next step S502.

  In step S502, the acoustic feature quantity related to the input voice is extracted by the feature quantity calculation processing module 402. Next, in step S503, the likelihood calculation processing module 403 reads the acoustic model for mounting position identification created in advance from the external storage device 104 via the acoustic model storage module 404, and the feature amount extracted in step S502. The acoustic likelihood is calculated using the read acoustic model.

  In step S504, the state display module 405 presents information regarding the suitability of the microphone mounting position using the acoustic likelihood calculated in step S503. Here, for example, a notification that generates a beep sound that increases in accordance with the magnitude of the likelihood, and a mode in which the transition of the acoustic likelihood is displayed in a graph on the information output unit 106 are included. Further, for a user who does not know the acoustic likelihood, it may be displayed that the microphone mounting position is appropriate when the acoustic likelihood exceeds a predetermined threshold. The user can determine whether or not the microphone is mounted at an appropriate position based on such display.

  In step S505, the information input processing module 406 receives feedback from the user through a button, a keyboard, or the like. If feedback for readjustment of the microphone mounting position is received, the process returns to step S501 to collect sound. When feedback indicating the determination of the position is received, the microphone mounting position is fixed and the process ends.

  FIG. 6 is a flowchart showing the flow of the speech recognition process in the present embodiment.

  This figure is mainly composed of three processes. Step S601, which is the first process, is a summary of the above-described microphone mounting position identification process, and after step S601, the microphone is mounted at the optimum sound collection position. As a result, the process moves to step S602. In step S <b> 602, the recognition acoustic model created by the recognition acoustic model creation module 204 is read from the external storage device 104 via the data storage module 202. In step S603, voice recognition is performed using the acoustic model read in the step 602 of the input voice from the microphone mounted in step S601, the result is output, and the process ends.

  As described above, according to the present embodiment, since the information regarding the acoustic likelihood is displayed based on the input voice before the voice recognition is performed, the operator has an opportunity to adjust the mounting position of the microphone based on the information. Is given.

(Embodiment 2)
In the first embodiment described above, one acoustic model for identifying the mounting position at the optimal sound collection position is created in advance, and the user is allowed to adjust the mounting position of the microphone based on this, but in this embodiment, An acoustic model when a microphone is mounted is designed for each of the optimum sound collection position and a plurality of positions around the optimum sound collection position, and a method of presenting the user with the direction in which the microphone mounting position should be adjusted based on this is shown. In the present embodiment, a case where a plurality of microphones are simultaneously attached to collect sound will be described. However, the present invention is not limited to this, and sound may be collected separately.

  Hereinafter, an outline of the microphone mounting position confirmation process in the present embodiment will be described.

  FIG. 9A is a diagram showing the periphery of a human neck, and A to E show examples of microphone mounting positions. Here, the position E is the optimum sound collection position. However, when the microphone is mounted without giving any instruction to the user, it is considered that there are many cases where the microphone is mounted at positions such as A, B, C, and D instead of E.

  In the present embodiment, in addition to creating an acoustic model based on voice data collected from a microphone mounted at the optimum sound collection position E, each position of A, B, C, D is mounted at that position. An acoustic model based on voice data collected from a microphone is created in advance. In the following, acoustic models corresponding to the positions A, B, C, D, and E are referred to as acoustic models a, b, c, d, and e, respectively.

  When a user actually wears a microphone and inputs sound, for each of the above acoustic models, the acoustic likelihood when using the acoustic model is calculated, and which acoustic model has the maximum acoustic likelihood. The position where the microphone is attached is estimated by examining whether or not it is.

  As a result of this estimation, if the mounting position is not the optimum E but A, B, C, or D, the direction from the position toward E is displayed to the user in that direction. Encourage the user to adjust the microphone position.

  The direction from the estimated wearing position to the optimum position E is, for example, an acoustic model having the maximum acoustic likelihood, and information on the direction from the wearing position estimated thereby to the optimum position E (that is, to be adjusted) An adjustment direction determination table in which the (direction) is described in association with each other is created in advance and can be determined by referring to the adjustment direction determination table. FIG. 9B shows an example of the structure of such an adjustment direction determination table. Thus, in the adjustment direction determination table, the acoustic models a, b, c, and d are described in correspondence with the direction from the mounting position corresponding to each acoustic model to the optimum sound collection position E.

  FIG. 7 is a diagram showing a module configuration of the speech recognition program involved in the process of creating the adjustment direction determination table in the present embodiment.

  The modules 701 to 704 are similar to the modules 201 to 204 in FIG. The difference from FIG. 2 is that a table creation module 705 is added. The table creation module 705 functions to create an adjustment direction determination table having a structure as shown in FIG. 9B, for example, and hold it in the external storage device 104 via the data storage module 702.

  FIG. 8 is a flowchart showing a flow of processing for creating an adjustment direction determination table in the present embodiment.

  First, in the processes of steps S801 to S803, acoustic models corresponding to a plurality of mounting positions deviating from the optimum sound collection position are created. These steps S801 to S803 correspond to the processing of steps S301 to S303 shown in FIG. However, steps S801 to S803 are performed independently for each mounting position, and an acoustic model for each mounting position is created. The plurality of acoustic models created here are all acoustic models for mounting position identification. When the acoustic model for each mounting position has been created, the process proceeds to the next step S804.

  In step S804, the table creation module 705 creates an adjustment direction determination table describing the direction from the corresponding microphone mounting position to the optimum sound collection position for each acoustic model. Specifically, referring to FIG. 9A, for example, A is located diagonally to the left of the optimum position E, so that the direction to be adjusted is diagonally upward to the right. Similarly, since B is located on the upper right side of the optimum position E, the direction to be adjusted is the lower left side. The adjustment direction determination table created in this way is stored in the external storage device 104 via the data storage module 702.

  Next, the microphone mounting position confirmation process in the present embodiment will be described with reference to the flowchart of FIG. The configuration of the speech recognition program module involved in the microphone attachment position confirmation process is the same as that shown in FIG. 4, and therefore the description thereof is omitted here, but the reference numerals are used.

  First, in step S1001, the voice input module 401 performs voice input from a microphone attached to the user. If there is no voice input, the state is maintained while the input state is maintained. If there is a voice input, the process proceeds to the next step S1002. In step S1002, the feature amount calculation processing module 402 extracts the acoustic feature amount of the input voice.

  Next, in step S1003, the likelihood calculation processing module 403 reads the acoustic models A, B, C, D, and E created in advance from the external storage device 104 via the acoustic model storage module 404, and extracts them in step S1002. An acoustic likelihood is calculated for each acoustic model with respect to the feature amount.

  Next, in step S1004, the acoustic likelihoods when using each acoustic model calculated in step S1003 are compared, and thereby the current microphone mounting position is estimated. For example, when the acoustic likelihood when the acoustic model E is used is larger than any other acoustic likelihood, it is estimated that the current microphone mounting position is at the optimum position E. In this case, there is no need to adjust the mounting position, so this process is terminated as it is. On the other hand, if the acoustic likelihood is not maximum when the acoustic model E is used, the current microphone mounting position is estimated to be any one of A, B, C, and D, and the process proceeds to step S1005.

  In step S1005, the information output unit 106 displays the direction in which the microphone should be adjusted with reference to the adjustment direction determination table stored in the external storage device 104. For example, when the likelihood of the acoustic model A is the maximum, “move right upward” is displayed on the information output unit 106. Since the direction in which the microphone mounting position is to be adjusted is presented in this way, the user can move the microphone mounting position without losing the adjustment direction. If the mounting position of the microphone moves as a result of the display process in step S1005, the process returns to step S1001 and the process is repeated.

(Embodiment 3)
In Embodiment 2 described above, the direction in which the microphone mounting position that can be presented to the user should be adjusted is limited to the directions defined by the adjustment direction determination table (for example, four directions). It is also possible to present an arbitrary direction without being constrained by.

  In this embodiment, the acoustic likelihood magnitudes of the four types of acoustic models a, b, c, and d (excluding the acoustic model e) calculated in the second embodiment and directions defined in the adjustment direction determination table are used. Determine the fine display direction. For example, the acoustic likelihood of each acoustic model is regarded as the magnitude of the vector, and the direction defined by the adjustment direction determination table for each acoustic model is regarded as the direction of the vector. Thereby, the direction presented as the direction to be adjusted can be represented by the sum of these vectors. This is formulated as follows.

That is, the display direction vector X to be obtained can be expressed by the sum of the direction vector E of the acoustic model i (i = a, b, c, d) multiplied by the acoustic likelihood P _i and the weighting coefficient ω _i .

  The result obtained in this way may be displayed on a display device such as the information output unit 106, or a direction such as an angle may be presented as a voice.

  It is also effective to present the distance to be moved in accordance with the size of the display direction vector X.

(Embodiment 4)
In the first to third embodiments described above, the microphone mounting position is adjusted based on the acoustic likelihood, but another evaluation function may be used instead of the acoustic likelihood.

  For example, there will be described a case where there are five acoustic models as in the second embodiment, and the acoustic likelihood determined from each model is used as the evaluation function. The processing flow is the same as in the first embodiment.

  As the position x where the microphone is actually mounted approaches the optimum sound collection position E, the acoustic likelihood when the acoustic model e is used for the feature quantity of the input voice by the microphone mounted at the position x increases, The acoustic likelihoods when the acoustic models a, b, c, and d are used can be expected to be small. This is expressed in the evaluation function as follows.

Here, P _i represents the acoustic likelihood of the input voice from the microphone mounted at the position x when the acoustic model i is used, and w _i represents the weighting factor of P _i .

(Embodiment 5)
By realizing the first and second embodiments together, it is possible to guide the mounting position of the microphone more efficiently. For example, the direction in which the microphone mounting position should be adjusted according to the second embodiment is first displayed for the mounted microphone, and the microphone approaches the optimum sound collection position while moving in the displayed direction. By switching to the beep sound according to the first embodiment or the acoustic likelihood display based on the screen display, it is possible to efficiently guide the mounting position of the microphone.

(Embodiment 6)
In Embodiments 1 to 5 described above, the method of precisely identifying the mounting position of the microphone and performing speech recognition using an acoustic model depending on the mounting position is shown, but in this embodiment, the mounting position of the microphone is shown. Shows a method of creating a sound model that does not depend strictly and performing speech recognition using it.

  FIG. 11 is a flowchart showing a flow of processing for creating an acoustic model that is robust against displacement of the microphone mounting position in the present embodiment. Here, a case where a plurality of microphones are simultaneously attached to collect sound will be described. However, the present invention is not limited to this, and sound may be collected separately.

  First, in step S1101, a plurality of microphones are mounted at and around the optimum sound collection position shown in the first embodiment. Next, in step S1102, all acoustic feature values are extracted from all input voice data, and in step S1103, all feature values are output to the external storage device 104.

  In step S1104, learning is performed using all the feature values recorded in the external storage device 104. This creates an acoustic model that is robust against displacement of the microphone mounting position. The created acoustic model is recorded in the external storage device 104 in step S1105.

  FIG. 12 is a flowchart showing the flow of speech recognition that is robust against the displacement of the microphone mounting position in the present embodiment.

  First, in step S1201, a voice input is received from a microphone attached to the user. In step S1202, feature amounts are extracted, and in step S1203, an acoustic model created according to the flowchart of FIG. 11 is read. In step S1204, speech recognition is performed based on the acoustic model read in step S1203 and the feature amount extracted in step S1202, and the recognition result is output in step S1205.

(Embodiment 7)
In the first to fifth embodiments, it has been described that the acoustic model created using only the voice input from the optimum sound collection position is used for actual voice recognition. However, the present invention is not limited to this. The speech recognition may be performed using an acoustic model that is robust against the displacement of the microphone mounting position described above.

(Embodiment 8)
Here, a specific example of the voice input unit 107 applicable to each of the above-described embodiments is shown.

  In FIG. 13, reference numeral 1301 denotes an example of a voice input device having a structure in which a plurality of microphones are arranged in one housing. In Embodiments 1 to 7 described above, it is assumed that a single microphone is used when performing voice input when the microphone is mounted. However, when voice input is performed, a voice input device such as 1301 is mounted, Performing speech recognition using the microphone with the highest likelihood as the input microphone at the time of speech recognition by independently calculating the likelihood of the speech input to each of the plurality of microphones and the acoustic model It becomes possible.

  On the other hand, reference numeral 1302 denotes a voice input device having a structure in which only one microphone for collecting sound is held inside and the microphone automatically travels within a predetermined range in the casing. A voice input device 1302 is mounted near the optimal sound collection position, and in the step of calculating the acoustic likelihood, the internal microphone automatically travels within a predetermined range in the housing, and the likelihood is calculated at each position. From the result of likelihood calculation over the entire movable range in the housing, the sound collecting microphone can be fixed at the position where the maximum likelihood is calculated in the housing, and then speech recognition can be performed.

(Other embodiments)
The embodiment of the present invention has been described in detail above. However, the present invention can take an embodiment as a system, apparatus, method, program, storage medium, or the like. In addition, the present invention may be applied to a system composed of a plurality of devices, or may be applied to an apparatus composed of a single device.

  In the present invention, a software program (a program corresponding to the flowchart shown in any of FIGS. 6 to 8) that realizes the functions of the above-described embodiments is directly or remotely supplied to a system or apparatus. This includes the case where the computer of the apparatus is also achieved by reading and executing the supplied program code. In that case, as long as it has the function of a program, the form does not need to be a program.

  Accordingly, since the functions of the present invention are implemented by computer, the program code installed in the computer also implements the present invention. That is, the scope of the claims of the present invention includes the computer program itself for realizing the functional processing of the present invention.

  In this case, the program may be in any form as long as it has a program function, such as an object code, a program executed by an interpreter, or script data supplied to the OS.

  As a recording medium for supplying the program, for example, flexible disk, hard disk, optical disk, magneto-optical disk, MO, CD-ROM, CD-R, CD-RW, magnetic tape, nonvolatile memory card, ROM, DVD (DVD-ROM, DVD-R).

  As another program supply method, a client computer browser is used to connect to an Internet homepage, and the computer program itself of the present invention or a compressed file including an automatic installation function is downloaded from the homepage to a recording medium such as a hard disk. Can also be supplied. It can also be realized by dividing the program code constituting the program of the present invention into a plurality of files and downloading each file from a different homepage. That is, a WWW server that allows a plurality of users to download a program file for realizing the functional processing of the present invention on a computer is also included in the claims of the present invention.

  In addition, the program of the present invention is encrypted, stored in a storage medium such as a CD-ROM, distributed to users, and key information for decryption is downloaded from a homepage via the Internet to users who have cleared predetermined conditions. It is also possible to execute the encrypted program by using the key information and install the program on a computer.

  In addition to the functions of the above-described embodiments being realized by the computer executing the read program, the OS running on the computer based on the instruction of the program is a part of the actual processing. Alternatively, the functions of the above-described embodiment can be realized by performing all of them and performing the processing.

  Furthermore, after the program read from the recording medium is written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, the function expansion board or The CPU or the like provided in the function expansion unit performs part or all of the actual processing, and the functions of the above-described embodiments are realized by the processing.

It is a block diagram which shows the structure of the speech recognition apparatus in embodiment. It is a figure which shows the structure of the module of the speech recognition program in connection with the acoustic model creation process in Embodiment 1. FIG. 3 is a flowchart showing a flow of acoustic model creation processing in the first embodiment. It is a figure which shows the structure of the module of the speech recognition program in connection with the microphone mounting position confirmation process in Embodiment 1. FIG. 6 is a flowchart illustrating a flow of microphone mounting position confirmation processing according to the first embodiment. 3 is a flowchart showing a flow of voice recognition processing in the first embodiment. It is a figure which shows the structure of the module of the speech recognition program in connection with the process which produces the adjustment direction determination table in Embodiment 2. FIG. 10 is a flowchart illustrating a flow of processing for creating an adjustment direction determination table in the second embodiment. It is a figure which shows the mounting position of the microphone in embodiment. It is a figure which shows the structural example of the adjustment direction determination table in Embodiment 2. FIG. 10 is a flowchart illustrating a microphone mounting position confirmation process according to the second embodiment. 14 is a flowchart showing a flow of processing for creating an acoustic model that is robust against a displacement of a microphone mounting position in the sixth embodiment. 14 is a flowchart showing a flow of speech recognition that is robust against displacement of the microphone mounting position in the sixth embodiment. It is a figure which shows the specific structural example of the audio | voice input part in Embodiment 13. FIG. It is a figure which shows the structural example of the audio | voice input part in embodiment.

Claims (14)

A voice recognition device that performs voice recognition using a microphone configured to collect weak voices that do not accompany the user's vocal cord vibrations when worn on a predetermined part of the user,
An acoustic model created based on speech collected from a speaker in a state where the microphone is mounted at an optimum sound collection position;
A calculation means for calculating an acoustic likelihood based on the acoustic model for a voice input by a user wearing the microphone;
Presenting means for presenting information about the suitability of the current mounting position of the microphone to the user using the acoustic likelihood calculated by the calculating means;
A speech recognition apparatus comprising: A voice recognition device that performs voice recognition using a microphone configured to collect weak voices that do not accompany the user's vocal cord vibrations when worn on a predetermined part of the user,
A plurality of acoustic models created for each position of the optimum sound collection position and a plurality of positions around the optimum sound collection position based on speech collected from a speaker with the microphone attached at that position;
Calculating means for calculating the acoustic likelihood based on one acoustic model of the plurality of acoustic models for the voice inputted by the user wearing the microphone; ,
Estimating means for estimating the current mounting position of the microphone based on the acoustic likelihood calculated by the calculating means;
When the wearing position estimated by the estimating means is not the optimum sound collecting position, the direction from the estimated wearing position to the optimum sound collecting position is defined as the direction in which the microphone wearing position should be adjusted. Presenting means to present to,
A speech recognition apparatus comprising:   The estimation means selects an acoustic model that maximizes the acoustic likelihood calculated by the calculation means, and determines the mounting position of the microphone at the time of collecting the voice that is the basis of the acoustic model. The speech recognition apparatus according to claim 2, wherein the speech recognition apparatus estimates the mounting position.   The presenting means refers to a table in which the plurality of acoustic models and a direction from the microphone mounting position toward the optimum sound collecting position at the time of collecting the voice that is the basis of each acoustic model are described in association with each other The voice recognition device according to claim 2, wherein a direction in which the mounting position of the microphone is to be adjusted is determined. A method for controlling a voice recognition device that performs voice recognition using a microphone configured to collect weak voice without vibration of the user's vocal cords by wearing it on a predetermined part of the user,
An input step of inputting voice from a user wearing the microphone;
A calculation step of calculating an acoustic likelihood of the voice input by the input step based on an acoustic model created from a voice collected from a speaker in a state where the microphone is mounted at an optimum sound collection position;
A presenting step of presenting information regarding the suitability of the current mounting position of the microphone to the user using the acoustic likelihood calculated in the calculating step;
A method for controlling a speech recognition apparatus, comprising: A method for controlling a voice recognition device that performs voice recognition using a microphone configured to collect weak voice without vibration of the user's vocal cords by wearing it on a predetermined part of the user,
An input step of inputting voice from a user wearing the microphone;
For each of the optimal sound collection position and a plurality of positions around it, one acoustic model of a plurality of acoustic models created from speech collected from a speaker with the microphone attached at that position A calculation step for calculating the acoustic likelihood of the speech input by the input step for all of the plurality of acoustic models;
An estimation step of estimating a current mounting position of the microphone based on the acoustic likelihood calculated by the calculation step;
When the wearing position estimated by the estimating step is not the optimum sound collecting position, the direction from the estimated wearing position to the optimum sound collecting position is set as the direction in which the microphone wearing position is to be adjusted. Presenting step to present to,
A method for controlling a speech recognition apparatus, comprising:   The estimation step selects an acoustic model having the maximum acoustic likelihood calculated by the calculation step, and determines the mounting position of the microphone at the time of collecting the voice that is the basis of the acoustic model, The method for controlling a speech recognition apparatus according to claim 6, wherein the position is estimated as a mounting position.   The presenting step refers to a table in which the plurality of acoustic models and the direction from the microphone mounting position toward the optimum sound collecting position in association with the sound that is the basis of each acoustic model are described in association with each other The method for controlling the speech recognition apparatus according to claim 6, wherein a direction in which the mounting position of the microphone is to be adjusted is determined.   A program for causing a computer to execute the method for controlling a speech recognition apparatus according to any one of claims 5 to 8.   A computer-readable storage medium storing the program according to claim 9. A voice recognition device that performs voice recognition using a microphone configured to collect weak voices that do not accompany the user's vocal cord vibrations when worn on a predetermined part of the user,
One acoustic model created for each of the optimum sound collection position and a plurality of positions around the optimum sound collection position based on the voice collected from the speaker with the microphone attached at that position;
Extracting means for extracting the feature amount of the voice input from the user wearing the microphone,
A speech recognition apparatus, wherein speech recognition is performed based on the feature amount extracted by the extraction unit and the acoustic model. A voice recognition method for performing voice recognition using a microphone configured to collect weak voice without vibration of the user's vocal cords by attaching to a predetermined part of the user,
An input step of inputting voice from a user wearing the microphone;
An extraction step of extracting a feature amount of the voice input by the input step;
A feature amount extracted by the extraction step, and an optimum sound collection position and each of a plurality of positions around it are created based on speech collected from a speaker with the microphone mounted at that position. A speech recognition step for performing speech recognition based on the acoustic model of
A speech recognition method comprising: In order to perform voice recognition using a microphone configured to collect weak voice without attaching the user's vocal cord vibration by wearing it on a predetermined part of the user,
An input step for inputting voice from a user wearing the microphone;
An extraction step for extracting a feature amount of the voice input in the input step;
A feature amount extracted by the extraction step, and an optimum sound collection position and each of a plurality of positions around it are created based on speech collected from a speaker with the microphone mounted at that position. A speech recognition step for performing speech recognition based on the acoustic model of
A program for running   A computer-readable storage medium storing the program according to claim 13.

Images