JP2001290495A - Device and method for speech recognition, and storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to be responded by authenticating a user himself (herself) only when the user himself (herself) makes a speech input. SOLUTION: Whether or not user's speech signals are inputted into a highly sensitive bone transmission microphone 105 is detected. User's featured values and speech featured values are extracted in a signal analysis section 109 based on the inputted speech signals. A user certifying section 106 computes a degree of similarity with the user's featured values while referring to the user's featured values beforehand stored in a RAM 102. When the user is not authenticated, the process stopts here. A speech recognition section 109 conducts speech recognition of the speech signals by referring to the speech featured values stored in the RAM 102. A speech recognition result obtained by the section 109 is converted into character information by a CPU 101 and the information is displayed in a character display area 401 and at the same time, a message requesting a next speech input is displayed for the user in an information display area 403.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device for controlling a device by voice and inputting voice to the device for information processing of voice, a control method thereof, and a storage medium.

[0002]

2. Description of the Related Art Conventionally, as a voice recognition device for controlling a device by voice or inputting voice to the device in order to process voice, there are a telephone with a voice recognition function, a car navigation, a personal computer and the like.

[0003] When security of personal information of a user is required in using these devices, it is desirable that the device having the voice recognition function can be operated only in response to the voice of the user.

Japanese Patent Laid-Open No. 8-186654 discloses a technique for responding to only the voice of the user (performing voice recognition).

In Japanese Patent Application Laid-Open No. Hei 8-186654, in a portable terminal device provided with a voice recognition input means, the voice recognition input means uses an input voice based on a personal voice feature amount specifying a user of the mobile terminal device. And a means for outputting the voice recognition data only when the input sound is determined to be the user's voice by the voice determining means. It is characterized by the following.

[0006]

However, in a device having a conventional voice recognition function, a user's own voice is recorded on a storage medium such as a tape, and data recorded on the storage medium (user's own voice) is recorded. If this is entered into this device, this device may respond. In addition, there is a possibility that voice recognition may not be performed correctly due to the influence of noise such as voices of other people other than the voice of the user himself, and these problems are not considered.

An object of the present invention is to solve the above problems and to provide a voice recognition apparatus which authenticates and responds to a user only when the user himself / herself inputs a voice, a control method thereof, and a storage medium.

[0008]

To achieve the object of the present invention, for example, the present invention has the following arrangement. That is, based on a voice signal input by a user, perform voice recognition and extraction of a user feature amount that specifies the user,
As a result of user authentication using the user feature, if the user is authenticated, a voice recognition device that responds using a voice recognition result, input means for inputting the voice signal using a bone conduction microphone, Voice feature extraction means for extracting a voice feature from an input voice signal; user feature extraction means for extracting the user feature from an input voice signal; and performing voice recognition using the voice feature. The apparatus includes a voice recognition unit, an authentication unit that authenticates a user by using the user characteristic amount, and a response unit that responds by using a voice recognition result when the user is authenticated.

[0009]

Embodiments of the present invention will be described below in detail with reference to the accompanying drawings.

[First Embodiment] In the present embodiment, each part in a speech recognition apparatus which determines a legitimate user by voice and outputs the voice as character information only when the legitimate user inputs the voice. Will be described.

FIG. 1 is a block diagram showing the inside of a speech recognition apparatus according to this embodiment.

Reference numeral 1 denotes the above-described speech recognition device.

Reference numeral 101 denotes a CPU which performs various processes described below based on various program codes stored in the ROM 103.

Reference numeral 102 denotes a RAM which stores a voice feature used when performing voice recognition in the voice recognition unit 109, a user feature used when performing user authentication in the user authentication unit 106, and various other control data and data. I do. Further, a work area used while the CPU 101 is executing the program code is provided.

A ROM 103 stores various program codes to be executed by the CPU 101. The display unit 1
Also, character codes for displaying characters are stored in the storage unit 10.

Reference numeral 105 denotes a high-sensitivity bone conduction microphone, which can be attached to the user's ear to detect vibrations inside the ear, including high-frequency components, with higher sensitivity than conventional microphones by a known technique. As a result, the user's voice can be detected without being greatly affected by the noise around the user. In addition, the high-sensitivity bone conduction microphone 105 performs acoustoelectric conversion of the detected vibration inside the ear.

Reference numeral 106 denotes a user authentication unit which performs user authentication by a method described later with reference to the user characteristic amount stored in the RAM 102 under the control of the CPU 101.

Reference numeral 107 denotes a signal amplifying unit for amplifying a user's voice signal input through the high-sensitivity bone conduction microphone 105.

Reference numeral 108 denotes a signal analyzer, and a signal amplifier 107
The pitch frequency and the power spectrum are calculated from the amplified audio signal by the method described later, and the above-described audio feature amount and user feature amount are extracted. Note that both cepstrum are used as the audio feature amount and the user feature amount in the present embodiment.

Reference numeral 109 denotes a voice recognition unit which performs voice recognition of a voice signal input by a user with reference to a voice feature stored in the RAM 102 under the control of the CPU 101. Note that, although DP matching is used for the speech recognition method, it is obvious that other known methods and techniques such as HMM may be used.

Reference numeral 110 denotes a display unit constituted by a CRT, a liquid crystal screen, or the like, and shows a result of voice recognition to a user. In the present embodiment, when the user who has input the voice is authenticated by the user authentication unit 106, the CPU 101 converts the result of the voice recognition by the voice recognition unit 109 into character information and displays the result on the display unit 110. FIG. 4 shows an example of a display screen when a result obtained by converting the speech recognition result by the speech recognition unit 109 as character information in the CPU 101 is displayed on the display unit 119.

Reference numeral 401 denotes a text display area for displaying a result obtained by converting a voice signal input by a legitimate user authenticated by the user authentication unit 106 as character information. It should be noted that Chinese characters are used as character information displayed in the character information area 401 in FIG. 3, which will be described later.

Reference numeral 402 denotes a cursor.

Reference numeral 403 denotes an information display area in which system messages from the voice recognition device 1 are displayed. For example, when the signal level of the audio signal input to the high-sensitivity bone conduction microphone 105 is equal to or lower than a predetermined level described later, this is an area for displaying a message such as "The voice is too low." The various messages are stored in the ROM 103 as character code data.

Reference numeral 404 denotes a user name display area.
An area 11 is used by the user to input a keyword described later at the time of authentication.

In the above-described display screen configuration, when all the voice signals input to the voice recognition device 1 are converted into characters and all are displayed in the text display area, the next voice is input to the information display area. Display a prompt message,
For example, in the same figure, "Please input voice" is displayed.

Returning to FIG. 1, reference numeral 111 denotes an operation unit which includes various switches and the like, and is used when various settings of the speech recognition apparatus 1 are performed.

Reference numeral 112 denotes an interface unit (hereinafter referred to as I /
F), a peripheral device such as a printer can be connected. For example, the character information output by the voice
The data can be output to a printer via F110 and printed on paper.

Reference numeral 113 denotes a bus connecting the above-described units.

Also, it is assumed that the voice feature amount and the user feature amount stored in the RAM 102 have been updated in advance with the latest training voice data of the user.

Next, FIG. 2 is a flowchart showing a flow of processing performed by each unit in the speech recognition apparatus 1 according to the present embodiment having the above-described configuration, and will be described.

In step S101, it is detected whether or not the user's voice signal has been input to the high-sensitivity bone conduction microphone 105. Then, in the high-sensitivity bone conduction microphone 105,
If an audio signal is detected, the next step is step S
The process proceeds to 102.

In step S102, the signal level of the audio signal input to the high sensitivity bone conduction microphone 105 is detected. If the signal level is equal to or higher than the reference level set by the operation unit 111, the process proceeds to the next step, step S103. Note that this reference level may be stored in the ROM 103, in which case it is not necessary to set the reference level in the operation unit 111. If the signal level of the audio signal is lower than the above-mentioned reference level, a message such as “the voice is too low” is displayed in the above-mentioned information display area 403.

In step S103, the audio signal input through the high-sensitivity bone conduction microphone 105 is amplified by the signal amplifying unit 108 to the target signal level set in the operation unit 111. In addition. This target signal level is stored in ROM1
03 may be stored, in which case setting of the target signal level in the operation unit 111 is not necessary.

In step S104, the signal amplifying unit 1
In 08, the signal analysis unit 109 extracts a user feature and a voice feature based on the voice signal amplified to the above-described predetermined target signal level by a method described later.

In step S 105, a message prompting the user to enter a keyword (eg, a user name) is displayed in the user name display area 404 in the information display area 403. On the other hand, under the control of the CPU 101, the user authentication unit 106 refers to the user feature stored in the RAM 102 in advance and calculates the similarity with the user feature extracted in step S104. For this similarity, a correlation coefficient is used. That is, a correlation coefficient between the user characteristic amount stored in the RAM 102 in advance and the user characteristic amount extracted in step S104 is calculated. The similarity is a similarity within the reference similarity set by the operation unit 111, and a keyword linked to the data of the user characteristic amount (the user indicated by the user characteristic amount is stored in the RAM 102 in advance. If the keyword input by the user using the operation unit 111 matches the keyword input in the user name display area 404, the user inputting the voice signal detected in step S101 is authenticated, and the process proceeds to step S106. Note that this reference similarity may be stored in the ROM 103, and at that time, setting of the reference similarity in the operation unit 111 is not necessary.

If the user has not been authenticated, the process ends here. Similarly, if the user is not authenticated, there is a possibility that the input method of the user's voice signal is wrong. Therefore, in order to give the user permission for inputting the voice signal a predetermined number of times, the setting in the operation unit 111 is performed. Alternatively, the process may return to step S101 the number of times that the input permission is given to the user. In this case, a message such as "Please enter again" is displayed in the information display area 403.

In step S106, the speech recognition unit 1
Reference numeral 09 performs voice recognition of a voice signal with reference to the voice feature amount stored in the RAM 102. The above-mentioned known method is used for the method of voice recognition. Note that the speech recognition processing in the speech recognition unit 109 is parallel to step S105.
Or you may go first.

In step S107, step S
At 106, the CPU 101 converts the speech recognition result by the speech recognition unit 109 into character information. This character information is information on how to read, and conversion from this reading to kanji is performed by using the kanji conversion table shown in FIG.
Do. This kanji conversion table is stored in the ROM 103. As a result, the converted Chinese characters are finally displayed in the text display area 401 as character information to be displayed in the text display area 401, and a message prompting the user to input the next voice is displayed in the information display area 4.
03 is displayed.

FIG. 3 shows the flow of a user's voice signal inside the voice recognition device 1. Note that the description of the figure is the same as that described above, and thus will be omitted.

FIG. 5 shows a flowchart of the signal analysis unit 108 for calculating the audio feature amount and the user feature amount from the audio signal, which will be described.

In step S501, the audio signal is subjected to fast Fourier transform (FFT). Since the processing in this step is to extract the frequency component of the audio signal, this step is not limited to FFT, and may be wavelet transform.

In step S502, the frequency components of the audio signal are analyzed. Specifically, step S50
In the frequency component of the audio signal extracted in step 1,
A histogram is generated for each frequency component (spectrum conversion).

In step S503, since the voice contains the pitch which is the fundamental frequency of the human vocal cords, the pitch is detected from the spectrum, and
Envelope information is also detected.

In step S504, step S5
Since the low-frequency component of the envelope information detected in 03 includes information on the vocal tract, which is one of the personal information of a person, the low-frequency component is extracted from the envelope information.

In step S505, step S
The vocal tract information of the user extracted in 504 is converted into a cepstrum, which is a characteristic parameter well representing personal information.

By performing the above-described processing method and the above-described processing method as a control method of the voice recognition device 1 therein,
It is possible to perform authentication of a legitimate user who does not depend on ambient noise and recognize voice of the legitimate user. Thereafter, it becomes possible to perform various processes using the present apparatus.

[Other Embodiments] Even if the above-described embodiment is applied to a system composed of a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), it is composed of one device. The present invention may be applied to an apparatus (for example, a copying machine, a facsimile machine, etc.).

An object of the above-described embodiment is to supply a storage medium (or a recording medium) in which a program code of software for realizing the functions of the above-described embodiment is recorded to a system or an apparatus, and to provide the system or the apparatus. It is needless to say that the present invention is also achieved when a computer (or a CPU or an MPU) reads and executes a program code stored in a storage medium. In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiment, and the storage medium storing the program code constitutes the above-described embodiment. By executing the program code read by the computer, not only the functions of the above-described embodiments are realized, but also an operating system (OS) running on the computer based on the instruction of the program code. It goes without saying that a case where some or all of the actual processing is performed and the functions of the above-described embodiments are realized by the processing is also included.

Further, after the program code read from the storage medium is written into a memory provided in a function expansion card inserted into the computer or a function expansion unit connected to the computer, the program code is read based on the instruction of the program code. Needless to say, the CPU included in the function expansion card or the function expansion unit performs part or all of the actual processing, and the processing realizes the functions of the above-described embodiments.

When the above-described embodiment is applied to the above-mentioned storage medium, the storage medium has been described previously (FIGS. 2 and 5).
The program code corresponding to the flowchart shown in FIG.

[0052]

[Effect of the Invention] Only when the user himself makes a voice input,
This has the effect of authenticating the user and responding.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the inside of a speech recognition apparatus according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating a flow of a process performed by the voice recognition device according to the first embodiment of the present invention.

FIG. 3 is a diagram showing a flow of a user's voice signal inside the voice recognition device.

FIG. 4 shows a result of speech recognition by a speech recognition unit 109 by a CPU.
The display unit 1 converts the information as character information in 101.
FIG. 10 is a diagram showing an example of a display screen displayed on the display 10;

FIG. 5 is a flowchart of a signal analysis unit 108;

FIG. 6 is a diagram showing a kanji conversion table.

Claims (15)

[Claims] 1. A speech recognition and extraction of a user feature for specifying the user based on a speech signal input by the user, and a result of user authentication using the user feature,
What is claimed is: 1. A voice recognition apparatus method for responding using a voice recognition result when a user is authenticated, comprising: an input unit configured to input the voice signal using a bone conduction microphone; and extracting a voice feature amount from the input voice signal. Voice feature extracting means for extracting, the user feature quantity extracting means for extracting the user feature quantity from the input voice signal, speech recognizing means for performing speech recognition using the voice feature quantity, and using the user feature quantity. A voice recognition device comprising: an authentication unit that authenticates a user by using the authentication unit; and a response unit that responds by using a voice recognition result when the user is authenticated. 2. The apparatus according to claim 1, further comprising a calculating unit configured to calculate a similarity between the user characteristic amount by the user characteristic amount extracting unit and a user characteristic amount prepared in advance. The voice recognition device according to claim 1, wherein the user is authenticated when the user is present in the device. 3. The speech recognition apparatus according to claim 2, wherein the user feature amount prepared in advance is updated with the latest training speech data of the user. 4. The speech recognition apparatus according to claim 1, wherein said speech recognition means performs speech recognition with reference to a user's speech feature quantity prepared in advance. 5. The speech recognition apparatus according to claim 4, wherein the speech feature amount of the user prepared in advance is updated with the latest training speech data of the user. 6. Based on a voice signal input by a user, voice recognition and extraction of a user feature specifying the user are performed, and as a result of user authentication using the user feature,
A voice recognition method for responding using a voice recognition result when a user is authenticated, comprising: an inputting step of inputting the voice signal to predetermined input means using a bone conduction microphone; A voice feature amount extraction step of extracting a feature amount; a user feature amount extraction step of extracting the user feature amount from an input voice signal; a voice recognition step of performing voice recognition using the voice feature amount; A voice recognition method comprising: an authentication step of authenticating a user using a feature amount; and a response step of responding using a voice recognition result when the user is authenticated. 7. A user characteristic amount extracted in the user characteristic amount extracting step and a calculating step of calculating a similarity between a user characteristic amount prepared in advance, and the similarity degree is calculated in the user authentication step. The voice recognition method according to claim 6, wherein the user is authenticated when the user is within a predetermined range. 8. The speech recognition method according to claim 7, wherein the user feature amount prepared in advance is updated with the latest training speech data of the user. 9. The speech recognition method according to claim 6, wherein in the speech recognition step, speech recognition is performed with reference to a user voice feature quantity prepared in advance. 10. The voice recognition method according to claim 9, wherein the voice feature amount of the user prepared in advance is updated with the latest training voice data of the user. 11. A computer that reads a voice signal based on a voice signal input by a user to perform voice recognition and extraction of a user characteristic value for specifying the user, and a result of user authentication using the user characteristic value. A storage medium that stores a program code that functions as a voice recognition device that responds by using a voice recognition result when a user is authenticated, wherein the voice signal is input to predetermined input means using a bone conduction microphone A program code of an input step, a program code of an audio feature amount extraction step of extracting an audio feature amount from an input audio signal, and a program code of a user feature amount extraction step of extracting the user characteristic amount from an input audio signal A program code for a voice recognition step of performing voice recognition using the voice feature amount; and a user code using the user feature amount. A program code authentication step of performing THE authentication, if the user is authenticated, the storage medium comprising: a program code of the response step for responding with a speech recognition result. 12. The program code of a calculating step of calculating a similarity between a user feature quantity extracted in the user feature quantity extraction and a user feature quantity prepared in advance, wherein the program code of the user authentication step is The storage medium according to claim 11, wherein the user is authenticated when the similarity is within a predetermined range. 13. The storage medium according to claim 12, wherein the user feature amount prepared in advance is updated with the latest training voice data of the user. 14. The storage medium according to claim 13, wherein in the voice recognition step, voice recognition is performed with reference to a voice feature amount of a user prepared in advance. 15. The storage medium according to claim 14, wherein the voice feature amount of the user prepared in advance is updated by the latest training voice data of the user.