JP2004212533A - Voice command adaptive equipment operating device, voice command adaptive equipment, program, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice command adaptive equipment operating device capable of providing excellent operability according to voice commands for many unspecified users. <P>SOLUTION: A parameter storage device 12 stores by users optimum parameters among voice recognition environmental parameters which are parameters for recognition precision improvement such as parameters in voice section extraction and parameters of sound models. A user is specified by a user specifying device 11. An operation history storage device 16 stores operation histories of voice command adaptive equipment by users. When a voice recognition error occurs, its cause is estimated from the operation histories to specify a voice recognition environmental parameter of the user to be adjusted and a parameter adjusting device 17 adjusts the parameter to automatically optimize voice recognition environmental parameters. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a voice command-compatible device operating device, a voice command-compatible device, a program, and a recording medium, and more particularly to operating a voice command-compatible device such as an office device that can be operated using voice recognition. The present invention relates to a voice command-compatible device operating device, a voice command-compatible device incorporating the device, a program incorporated in the device, and a recording medium.
[0002]
[Prior art]
As can be seen from the contents of the Amendment to the Rehabilitation Act 508, which was enforced in the United States last June, the development of office equipment that enables people with physical disabilities to operate as well as healthy persons has recently been required. .
[0003]
In particular, operations such as recent office equipment that use a variety of information displayed on a flat LCD touch panel according to the scene and virtual buttons as clues to select abundant functions and set modes The system is a major barrier, especially for the visually impaired.
[0004]
As a promising means to remove this barrier, a voice interface using voice guidance output or voice command input can be considered.
[0005]
In the device operation using voice commands, for software that is used by a specific individual in a specific environment, such as software on a personal computer, the user is required to utter a sample voice of tens to hundreds of words in advance to adjust the parameters of the acoustic model. In many cases, recognition accuracy is improved by performing learning called speaker adaptation for tuning. However, for devices that are used by a large number of users, such as office equipment, the speaker adaptation process that requires a considerable amount of time to register sample voices must be performed for each user in front of the actual equipment. Is difficult.
[0006]
Therefore, in general, an acoustic model suitable for an unspecified speaker capable of recognizing the voices of many speakers is used. In this case, some speakers exhibit extremely low recognition accuracy. there's a possibility that. As described above, it is impossible to completely eliminate a voice recognition error by inputting a voice command, and depending on a speaker, the recognition accuracy may be extremely low.
[0007]
The cause is
(1) The utterance start timing is inappropriate, and the beginning of the voice section is not properly extracted.
(2) A somewhat long breath or pause is mixed in the command, and the end of the voice section is not properly extracted.
(3) the voice quality of the speaker does not match the parameters of the acoustic model used;
Such causes are possible.
[0008]
[Problems to be solved by the invention]
The above-mentioned causes (1) and (2) are causes that occur in the voice section cutout processing performed as preprocessing of voice recognition. In this voice section cutout processing, appropriate parameters differ depending on the utterance style of the speaker. . Although the above causes (1) and (2) may be improved by training to some extent, there are cases where it is difficult for the user to respond due to some kind of obstacle. Therefore, it is necessary to appropriately select parameters at the time of speech section extraction. However, there is no device that changes (improves by adjusting) these parameters at present. For the above-mentioned cause (3), it is essential to improve the recognition accuracy by adjusting the parameters of the acoustic model.
[0009]
The present invention has been made in view of the above circumstances, and automatically adjusts parameters for improving recognition accuracy, such as parameters at the time of speech section extraction and acoustic model parameters, for each user. , A voice command-compatible device operating device, a voice command-compatible device incorporating the device, and a voice command-compatible device capable of providing operability with good voice commands to an unspecified number of users. It is an object of the present invention to provide a program to be executed and a computer-readable recording medium on which the program is recorded.
[0010]
[Means for Solving the Problems]
The invention according to claim 1 is a voice command-compatible device operating device that operates a voice command-compatible device having a voice recognition device that recognizes various commands input by voice, wherein a user specifying unit that specifies a user; A parameter storage means for storing an optimum voice recognition environment parameter for each user; an operation history storage means for storing an operation history of the voice command corresponding device by each user; and a voice of the user according to the operation history. Parameter optimization means for optimizing the recognition environment parameters.
[0011]
According to a second aspect of the present invention, in the first aspect of the present invention, the user identifying unit identifies the user by inputting an ID number by a button on the voice command compatible device such as a numeric keypad. And a means for identifying a user by registering and collating the user name using voice recognition, and a special means such as a magnetic card, a non-contact type IC card, a PHS, a mobile phone, and a personal wireless terminal. And at least one of specifying means for specifying a user by using a device / apparatus.
[0012]
According to a third aspect of the present invention, in the first or second aspect of the invention, the parameter optimizing means can adjust at least one of a timeout time, a pause allowable time, and a probability parameter of an acoustic model as the speech recognition environment parameter. The timeout time is a parameter that limits the time delay from when an operation for instructing the start of voice input such as pressing a button is performed until the actual utterance starts, and the pause allowable time is set in the middle of a command. It is characterized in that it is a parameter for limiting the time such as a breath or a pause to be inserted.
[0013]
According to a fourth aspect of the present invention, in the first aspect of the present invention, when a speech recognition error occurs, the parameter optimizing means estimates a cause of the error and specifies a parameter to be adjusted. It is characterized by having parameter specifying means.
[0014]
According to a fifth aspect of the present invention, in the fourth aspect of the invention, the parameter optimizing means adjusts a parameter related to a factor estimated as a cause of the speech recognition error so as to reduce the erroneous recognition. It is characterized by having.
[0015]
A sixth aspect of the present invention is a voice command compatible device incorporating the voice command compatible device operation device according to any one of the first to fifth aspects.
[0016]
According to a seventh aspect of the present invention, there is provided a program for causing a computer to function as the voice command-compatible device operating device according to any one of the first to fifth aspects.
[0017]
The invention according to claim 8 is a computer-readable recording medium on which the program according to claim 7 is recorded.
[0018]
BEST MODE FOR CARRYING OUT THE INVENTION
The voice command operation device according to the present invention is a device having a voice recognition device that recognizes various commands input by voice, that is, a device such as an office device that can be operated using voice recognition (in the present specification). This is an operation device for operating a voice command compatible device, and is a device useful for improving accessibility of the voice command compatible device. The voice command-compatible device operation device (often referred to as an operation device) according to the present invention includes a user identification unit, a parameter storage unit, an operation history storage unit, and a parameter optimization unit, which will be described later. Adjustment of parameters for improving recognition accuracy, such as parameters at the time of section extraction and acoustic model parameters, is automatically performed for each user, and as a result, a good voice command for an unspecified number of users Operability can be provided. The voice command compatible device according to the present invention is a device such as an office device that can be operated using voice recognition, and the operability can be improved by incorporating the operating device according to the present invention. It is possible.
[0019]
The parameter storage means stores the optimum voice recognition environment parameters for each user, and the operation history storage means stores the operation history of each user's voice command compatible equipment (such as voice command compatible office equipment). Further, the parameter optimizing means automatically optimizes the user's voice recognition environment parameters according to the operation history. Since the parameter optimizing means optimizes the voice recognition environment parameters according to the operation history, it can be said that the parameter optimizing means has a parameter learning function. By these means, it is possible to provide an operability with a good voice command to an unspecified number of users. For example, it is possible to provide an office device that can be operated even by a visually impaired person.
[0020]
The user specifying means is a means for specifying a user, and preferably has at least one of the following three specifying means. The first specifying means is a means for specifying a user by inputting an ID number by a button on the voice command compatible device such as a numeric keypad. The second specifying means is a means for specifying a user by registering and collating the user name using voice recognition. The third specifying means is a means for specifying a user by using a special device such as a magnetic card, a non-contact type IC card, a PHS, a mobile phone, or a personal wireless terminal. With this specifying means, it is possible to provide a framework for preparing an optimum voice recognition environment for each user in a voice command compatible device.
[0021]
Further, in the parameter optimizing means according to one embodiment of the present invention, at least one of a timeout time, a pause allowable time, and a probability parameter of an acoustic model can be adjusted as the speech recognition environment parameter. Here, the timeout time is a parameter that limits the time delay from when an operation for instructing the start of voice input, such as pressing a button, is performed to when an actual utterance starts. In addition, the pause allowable time is a parameter for limiting a time such as a breath or a pause inserted in the middle of a command. By providing these adjustable parameters in order to prepare an optimal speech recognition environment for each user, recognition accuracy is improved and erroneous recognition is reduced.
[0022]
In addition, it is preferable that the parameter optimizing means include an adjustment parameter specifying means for estimating a cause of a speech recognition error and specifying a parameter to be adjusted when a speech recognition error occurs. With this adjustment parameter specifying means, it is possible to provide a parameter adjustment framework for preparing an optimum voice recognition environment for each user in a voice command compatible device.
[0023]
Further, the parameter optimizing means includes not only the adjustment parameter specifying means but also a parameter adjusting means for automatically adjusting a parameter related to a factor estimated as a cause of a speech recognition error so as to reduce erroneous recognition. It is preferable to do so. By providing an adjustment method by the parameter adjustment means in order to prepare an optimum speech recognition environment for each user, recognition accuracy is improved and erroneous recognition is reduced.
[0024]
Further, as another embodiment of the present invention, a program for causing a computer to function as the operating device according to any of the above-described modes (for example, a program for executing a procedure of a process described later with reference to FIG. 1), A form as a computer-readable recording medium on which the program is recorded can also be adopted. The program is a program that is incorporated in the voice command-compatible device and that is to be controlled by a control arithmetic processing unit (represented as a computer) of the voice command-compatible device so as to realize the function according to the present invention. . Specific examples of the recording medium include a CD-ROM, a magneto-optical disk, a DVD-ROM, an FD, a flash memory, a memory stick, and various other ROMs and RAMs. The functions of the operating device are executed by a computer (including not only a general-purpose computer but also the above-described arithmetic processing unit (CPU, MPU, etc.) incorporated in the device), and a program for realizing this function is recorded. By realizing this function, it is easy to realize this function. Then, the recording medium as described above is mounted on the operation device or the voice command compatible device, and the program is read out by the operation device or the voice command compatible device, or the program is stored in a recording medium provided in the operation device or the voice command compatible device. By reading the data as needed, the function of operating the device corresponding to the voice command according to the present invention can be executed.
[0025]
FIG. 1 is a block diagram showing an example of a voice command compatible device according to an embodiment of the present invention. In the drawing, reference numeral 1 denotes a configuration example of a copy machine with a voice command function. A voice command compatible device according to an embodiment of the present invention (often referred to as the device) will be described using the copy machine 1 with a voice command function of FIG. 1 as an example.
[0026]
The copying machine 1 includes a user identification device 11, a parameter storage device 12 as a parameter storage unit, a microphone (hereinafter referred to as a microphone) 13, an A / D converter 14, a voice recognition device, in addition to the functions 10 of the copying machine body. 15, an operation history storage device 16 as an operation history storage device, and a parameter optimization device (shown as a parameter adjustment device 17 that performs its main function) as a parameter optimization device.
[0027]
The parameter storage device 12 is a storage device for storing optimum voice recognition environment parameters for each user, and the operation history storage device 16 is a storage device for storing operation history of the copy machine 1 by each user. It is. The parameter optimizing device having the parameter adjusting device 17 is a device that performs an operation to optimize the voice recognition environment parameter of the user according to the stored operation history. In the present invention, when a speech recognition error occurs according to the operation history stored in the operation history storage device 16, the cause is estimated and the speech recognition environment parameter to be adjusted by the user is specified. By adjusting the parameters by the parameter adjusting device 17, the speech recognition environment parameters are automatically optimized.
[0028]
The voice recognition device 15 performs voice section extraction means 23 for extracting a voice section from the voice data input via the microphone 13 and the A / D converter 14, and performs acoustic analysis on each of the cut voice sections. An acoustic analysis unit 24, a speech recognition unit 25 for recognizing speech based on data after the acoustic analysis, and a storage area (also referred to as a work area) 20 for temporarily storing speech-recognition-related parameters by storing or recalling the parameters. Shall be. Here, the voice recognition related parameters are parameters called from the parameter storage device 12 from the voice recognition environment parameters of the target user (here, the user 3), and the section extraction related parameters 21 used for voice section cutout. And acoustic model parameters 22 for recognizing speech from the acoustic analysis result.
[0029]
The processing procedure in the copy machine 1 is as follows according to the device configuration as described above. First, at the start of use of the device, the user uses the user identification device 11 to transmit the user ID to the device. This is performed, for example, by passing the magnetic card C through a card reader or the like.
[0030]
After confirming the user ID, the specified speech recognition-related parameters for the user are loaded from the speech recognition-related parameter storage device 12 into the work area 20. Thereafter, the user executes the operation of the device by key input of the copier main body or voice input via the copier main body or the microphone 13 and the A / D converter 14. The A / D converter 14 converts the information on all operations executed during the series of operations, that is, information on the timing of pressing each button and all the sound information input from the microphone 13 into digital data. The data is stored in the operation history storage device 16.
[0031]
FIG. 2 is a diagram illustrating an example of a time chart of the audio cutout processing.
First, with reference to FIG. 2, a description will be given of the cause of the failure of the operation by the voice command. The following three major causes (A) to (C) can be considered as causes of failure.
[0032]
(A) When the utterance start timing is delayed, the time-out process performed in the voice segment cutout process (when a voice segment is not detected even after a certain time (time-out time To) has elapsed after the voice input start button has been pressed). Interrupted the process), so the audio capture itself failed.
(B) The recognition process failed because the start and end of the voice uttered by the user could not be detected correctly.
(C) The recognition process failed because the voice quality of the user did not match the parameters of the acoustic model.
[0033]
The timeout process (A) is a process necessary for providing a comfortable voice input environment. Otherwise, if you accidentally press the voice input start button, you must either explicitly cancel it, or the voice input process will not be completed until some sound is input, so it will not be affected by external noise other than voice. The possibility that the device will malfunction due to the reaction increases. For this reason, the default value is set somewhat short (several hundred milliseconds), but it may be difficult for some users to speak within a predetermined time-out period To (due to some kind of trouble or the like).
[0034]
Therefore, regarding the failure of (A), the voice data stored as the operation history is analyzed, and when it is detected that some utterance has been made within a relatively short time after the occurrence of the timeout, the user of the user is analyzed. Is increased at a fixed rate. However, if the increase in the timeout time To is allowed without any limitation, the above-described problem occurs. Therefore, an appropriate upper limit (about 1 to 2 seconds) is actually set.
[0035]
On the other hand, when an erroneous recognition of a voice command such as (B) or (C) occurs, the user utters the voice command again, or performs the same operation again by another operation means such as a key operation. become. Therefore, if the operation is performed again without determining the mode selected according to the recognition result of the voice command input, it can be assumed that the recognition result of the voice command input was incorrect ( This is called a false recognition hypothesis). In this case, erroneous recognition may occur consecutively a plurality of times. In this case, one of the erroneous recognitions (for example, the first erroneous recognition) is set as the erroneous recognition hypothesis. Further, when the user determines an operation that has been redone after the false recognition hypothesis has been established, it can be assumed that the voice command corresponding to the operation is the correct answer to the previously falsely recognized voice command (this is the correct answer). Call it a hypothesis).
[0036]
However, there is no guarantee that the above-mentioned hypothesis is always correct because the content that the user wants to execute may have been suddenly changed after inputting the voice command. Therefore, when the pair of the misrecognition hypothesis and the correct answer hypothesis as described above is determined, the device (the copy machine 1 in this case) instructs the user that the voice command that has failed to first input is the correct answer hypothesis. Outputs a message to confirm that the message is valid and asks the user to confirm.
[0037]
The voice command data that failed to be input earlier is extracted from the voice data stored as the operation history and the data after the time when the voice input start button etc. indicating the start of the input was pressed, and used for actual voice recognition. By performing the used voice segment extraction processing, it is possible to faithfully reproduce. At this time, in the case (B) in which the detection of the start and end of the voice section has failed, the playback sound of the voice command data extracted here differs from the one intended by the user (one of the beginning of the sentence or the end of the sentence). Part is missing). On the other hand, in the case of (C), in which the voice section extraction succeeds but the subsequent recognition processing fails, the voice extracted here should be the same as that intended by the user.
[0038]
Therefore, in order to distinguish between the two, a message such as "Is the command input first" double-sided copy (this part is the reproduction sound of the real voice of the user extracted in the above processing) "is output. Then, a message such as "Is the correct answer of the previously input command" double-sided copy (this part is the correct answer hypothesis) "" is output. If the answer to the second question is YES, the cause of the erroneous recognition at that time and the pair of the erroneously recognized voice command data and its correct answer are determined. If the answer to the second question is NO, the correct answer cannot be specified, and the subsequent processing is not performed.
[0039]
If the first question is NO, that is, if the detection of the start and end of the voice section has failed, the following two causes can be considered.
(A) An unexpected sound, such as a sudden noise, was input immediately before or immediately after a voice.
(B) There was an unexpectedly long pause during command utterance, at which point the utterance was considered completed.
[0040]
The problem (a) is an essential problem that is difficult to deal with with the current speech recognition technology. However, the problem of the pause section of (b) is reduced to some extent by increasing the pause allowable time Tp. It is possible to avoid. However, in this case, even though the utterance is actually completed, the end cannot be determined unless the user waits for that time, so that the response to the voice command is deteriorated.
[0041]
For this reason, the default value is set to be somewhat short (several hundred milliseconds). However, depending on the user, it may be unavoidable that a pause longer than a predetermined allowable pause time is inserted (due to some kind of obstacle). .
[0042]
Therefore, for the voice data stored as the operation history, the pause allowable time Tp is temporarily increased at a fixed rate, and then the voice section is re-cut out, and the voice recognition result for the data is the correct hypothesis. Only when they match, the actual pause allowable time Tp is replaced with the (temporarily increased) value. However, if the increase of the pause allowable time Tp is allowed without any limitation, the above-mentioned problem occurs. Therefore, an appropriate upper limit (about 1 to 2 seconds) is actually set.
[0043]
One last reason, that is, if it is considered that the recognition process has failed because the voice quality of the user does not match the parameters of the acoustic model, the acoustic model using the erroneously recognized speech data and the correct answer information is used. Carry out speaker adaptation. As the acoustic model to be subjected to speaker adaptation, the default speaker-independent model is used for the speaker who performs this speaker adaptation for the first time, and the speaker already has its own adapted acoustic model. Use that model for existing speakers.
[0044]
This speaker adaptation processing is performed using a speaker adaptation algorithm suitable for adaptation processing with a small number of samples, such as a widely known MAP (maximum posterior probability) estimation method.
[0045]
The parameter changed in such a series of processing is updated from the work area 20 to the parameter storage device 12 for each speaker again each time the parameter is changed or when the series of operations is completed.
[0046]
【The invention's effect】
ADVANTAGE OF THE INVENTION According to this invention, the parameter adjustment for the improvement of recognition accuracy, such as the parameter at the time of speech area extraction and the parameter of an acoustic model, is automatically performed for every user, Operability by a good voice command can be provided.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating an example of a voice command compatible device according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating an example of a time chart of a voice cutout process.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Copy machine, 10 ... Copy machine main body function, 11 ... User identification device, 12 ... Parameter storage device, 13 ... Microphone, 14 ... A / D converter, 15 ... Voice recognition device, 16 ... Operation history storage device, 17: parameter adjustment device, 20: storage area (work area), 21: section extraction related parameters, 22: acoustic model parameters, 23: voice section cutout means, 24: acoustic analysis means, 25: voice recognition means.

Claims (8)

A voice command-compatible device operating device for operating a voice command-compatible device having a voice recognition device for recognizing various commands input by voice, comprising: a user specifying unit for specifying a user; Parameter storage means for storing voice recognition environment parameters; operation history storage means for storing operation histories of the voice command-compatible device by each user; and optimizing the user's voice recognition environment parameters according to the operation histories. And a parameter optimizing means. The user specifying means is a specifying means for specifying a user by inputting an ID number by a button on the voice command compatible device such as a numeric keypad, and a user name is registered using voice recognition. A means for specifying a user by collating beforehand, and a means for specifying a user by using a special device such as a magnetic card, a non-contact type IC card, a PHS, a mobile phone, or a personal wireless terminal. 2. The apparatus according to claim 1, further comprising at least one of the following. The parameter optimizing means is capable of adjusting at least one of a timeout time, a pause allowable time, and a probability parameter of an acoustic model as the speech recognition environment parameter, and the timeout time is used to start a speech input such as pressing a button. The parameter is used to limit the time delay from when the instructing operation is performed to the start of the actual utterance, and the pause allowable time is a parameter that limits the time such as breathing and pause inserted in the middle of the command. The voice command-compatible device operating device according to claim 1 or 2, wherein: 4. The apparatus according to claim 1, wherein said parameter optimizing means has an adjustment parameter specifying means for estimating a cause of the occurrence of a speech recognition error and specifying a parameter to be adjusted. Voice command compatible device operation device. 5. The apparatus according to claim 4, wherein said parameter optimizing means includes a parameter adjusting means for adjusting a parameter associated with a factor estimated as a cause of the speech recognition error so as to reduce the erroneous recognition. Equipment operation device. A voice command compatible device incorporating the voice command compatible device operation device according to claim 1. A program for causing a computer to function as the voice command-compatible device operating device according to claim 1. A computer-readable recording medium on which the program according to claim 7 is recorded.

Images