JP2007017840A - Speech authentication device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech authentication device capable of improving accuracy of speech authentication. <P>SOLUTION: A feature parameter of uttered speech of a person collected by a first microphone for collecting air-borne sound of the uttered speech of the person and a second microphone for collecting body transmission sound of the uttered speech of the person is extracted and stored beforehand, and a feature parameter is extracted from the uttered speech of the person collected by the first and second microphones and is verified with the stored feature parameter to authenticate whether or not the person is qualified as a user based on this comparison results. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a voice authentication device that authenticates whether a user is an appropriate user by comparing a voice uttered by a user with a voice stored in advance.

  Speak words such as a predetermined user name, analyze the user's speech and store it as a feature parameter, analyze the speech uttered by the user during authentication, and extract the feature parameter In addition, a voice authentication apparatus is known in which matching or mismatching is determined by comparing with stored feature parameters (see, for example, Patent Document 1).

Prior art documents related to the invention of this application include the following.
JP 2004-170522 A

  However, since the conventional voice authentication apparatus described above collects the air conduction sound and performs authentication, there is a problem that environmental noise is likely to be mixed into the user's voice and the authentication accuracy cannot be improved.

  Extracting feature parameters from the speech of a person collected in advance by a first microphone that collects the air conduction sound of the person's speech and a second microphone that collects the body conduction sound of the person's speech The feature parameters are extracted from the speech of the person collected by the first microphone and the second microphone, compared with each stored feature parameter and each microphone, and the user is based on the comparison result. It certifies whether or not it is eligible.

  According to the present invention, voice authentication accuracy can be improved.

  An embodiment in which the voice authentication device of the present invention is applied to a vehicle will be described. The voice authentication device of the present invention is not limited to vehicles.

  FIG. 1 is a diagram showing a configuration of an embodiment. A non-contact type microphone (hereinafter simply referred to as a microphone) 1 is installed, for example, on an instrument panel around the driver's seat, and collects the air conduction sound of the driver's speech. For example, a condenser microphone can be used as the non-contact type microphone 1. Since the non-contact type microphone 1 collects air conduction sound, environmental noise is mixed in the voice of the driver.

  The contact-type microphone 2 is generally called a bone conduction microphone or a meat conduction microphone, and collects the body conduction sound of the user's uttered voice by contacting the user's head, for example. The contact-type microphone 2 is used by a user wearing on the body using a fixing device or being embedded in a part that comes into contact with the body, such as a headrest.

  The frequency band that can be picked up by the microphone differs between the non-contact type and the contact type. Since high-frequency information is likely to be included in the air conduction sound, the non-contact type microphone 1 uses a microphone capable of collecting sound with emphasis on high-frequency information. On the other hand, since the low-frequency information is likely to be included in the body conduction sound, the contact microphone 2 sensitive to the low-frequency information is used. In addition, the high frequency information of the sound collected by the microphone having excellent low frequency characteristics may improve the authentication accuracy if it is not parameterized by filtering.

  The microphone amplifier 3 separately amplifies the sounds collected by the non-contact type microphone 1 and the contact type microphone 2, and the AD converter 4 analog sounds of the non-contact type microphone 1 and the contact type microphone 2 amplified by the microphone amplifier 3. Are individually converted into digital audio.

  The arithmetic unit 5 includes a microcomputer, a memory, etc., and analyzes the characteristics of the user's uttered voice and compares them with the user's characteristic parameters stored in the storage device 6 to authenticate whether the user is an appropriate user. I do. In the parameter analysis, information such as voice pitch, formant frequency, spectrum, cepstrum, speech speed, prosody, spectral regression, power, intonation, utterance content, etc. is used alone or in combination.

  The storage device 6 asks a user to speak a predetermined name in advance, records and analyzes the voice, extracts a characteristic parameter, and stores it for each user. In the voice authentication, the user's feature parameter stored in advance in the storage device 6 is compared with the feature parameter of the user's utterance voice input at the time of authentication. The characteristic parameters are obtained by performing processing such as frequency analysis, cepstrum analysis, pitch extraction, etc. on the speech signal, extracting parameters related to individuality included in the speech signal, and vectorizing them. FIG. 2 shows an example of creating a five-dimensional vector, that is, a feature parameter, using pitch information and formant information extracted from frequency analysis.

  FIG. 3 is a flowchart showing the speaker authentication operation of the embodiment, and FIG. 4 is a diagram showing the speaker authentication principle of the embodiment. The operation of one embodiment will be described with reference to these drawings. In step 1, initialization processing is performed, and feature parameters relating to the user's voice collected in advance from the storage device 6 are read and expanded in the memory space. In step 2, the presence or absence of a voice input by the user is detected. If voice is input from the non-contact microphone 1 and the contact microphone 2, the process proceeds to step 3.

  In step 3, feature parameters are extracted from the voices input from the non-contact microphone 1 and the contact microphone 2 by the method described above. In the subsequent step 4, the feature parameter extracted for each microphone is compared with a feature parameter stored in advance. That is, the feature parameter extracted from the input sound of the non-contact type microphone 1 is compared with the characteristic parameter recorded in advance by recording with the non-contact type microphone 1, and the characteristic parameter extracted from the input sound of the contact type microphone 2 is Compared with the characteristic parameters recorded in advance by recording with the contact microphone 2.

  As a feature parameter comparison method, for example, there is a method of comparing Euclidean distances between two feature parameters. FIG. 5 shows an example in which three speaker authentication is performed using a two-dimensional vector (feature parameter). If the input signal is in the range from C1 to C3, the speaker authentication is successful, that is, the user is qualified. Note that the accuracy of speaker authentication can be controlled by changing the range of C1. In order to increase the accuracy of authentication and make it strict, the range of C1 may be reduced.

  The authentication result is output for each microphone. When a matching result of matching is obtained, the user is qualified as “1”, and when a matching result of mismatching is obtained, the user is uncertainly “0”. In step 5, the logical product of the authentication results for each microphone is calculated, and the final authentication result is output. That is, it is assumed that the user is qualified only when the authentication result by the non-contact microphone 1 is qualified “1” and the authentication result by the contact microphone 1 is qualified “1”. Note that the range of authentication can be expanded by taking the logical sum of the authentication results for each microphone to obtain the final authentication result.

  Thus, according to one embodiment, the non-contact type microphone that collects the air conduction sound of the person's uttered voice and the contact type microphone that collects the body conduction sound of the person's uttered voice are collected in advance. Feature parameters are extracted and stored from the uttered speech of the person who has made the sound, extracted from the utterance speech of the person collected by the non-contact microphone and the contact microphone, and the stored feature parameters and each microphone The result of the comparison and verification was used to authenticate whether or not the user was eligible. Based on the result of the comparison and verification, the amount of information used for authentication was increased, and it was obtained from the body conduction sound that contained less environmental noise. The voice authentication accuracy can be improved by using the information.

  In addition, according to the embodiment, when matching results are obtained in both the non-contact type microphone and the contact type microphone, the user is authenticated as being qualified. The voice authentication accuracy can be improved by using the information obtained from the body conduction sound that increases in volume and has little environmental noise.

  Furthermore, according to the embodiment, when a matching result is obtained in either one of the non-contact type microphone and the contact type microphone, it is authenticated that the user is qualified. The voice authentication accuracy is lower than the case where the matching result is obtained in, but the voice authentication accuracy is higher than that of the conventional voice authentication device by using the information obtained from the body conduction sound with less environmental noise. Can be improved.

<< Modification of Embodiment of Invention >>
In the above-described embodiment, an example in which the feature parameter of the uttered voice at the time of authentication is compared and collated with the feature parameter stored in advance for each microphone is shown. However, the feature parameter of the two microphones is integrated and collated. May be.

  FIG. 6 shows a modified speaker authentication method. A user's name determined in advance is uttered in advance, and voice is recorded using the non-contact microphone 1 and the contact microphone 2. The voice recorded by the non-contact type microphone 1 and the voice recorded by the contact type microphone 2 are separately analyzed by the method described above, and feature parameters are extracted. Then, a part of the characteristic parameters of the non-contact microphone 1 and a part of the characteristic parameters of the contact microphone 2 are brought together and integrated into one characteristic parameter, and stored in the storage device 6 as a user characteristic parameter.

  At the time of authentication, the user's predetermined name is uttered, and the uttered voice is recorded by the non-contact microphone 1 and the contact microphone 2. The voice recorded by the non-contact type microphone 1 and the voice recorded by the contact type microphone 2 are separately analyzed by the method described above, and feature parameters are extracted. Then, a part of the characteristic parameters of the non-contact type microphone 1 and a part of the characteristic parameters of the contact type microphone 2 are brought together and integrated into one characteristic parameter, and compared with the characteristic parameter stored in the storage device 6. . If a matching result is obtained, the user is qualified, and if a mismatching result is obtained, the user is inaccurate.

  As described above, according to the modification of the embodiment, the non-contact microphone that collects the air conduction sound of the person's uttered voice and the contact microphone that collects the body conduction sound of the person's uttered voice in advance. The feature parameters are extracted from the uttered voice of the person who collected the sound and the feature parameters of the non-contact microphone and the feature parameters of the contact microphone are integrated to generate and store the feature parameters. In addition, the feature parameters are extracted from the speech uttered by the person collected by the non-contact microphone and the contact microphone, and the feature parameters of both microphones are integrated, compared with the stored feature parameters of the integrated results, and matched. When the verification result is obtained, it is authenticated that the user is qualified as a user, so that the amount of information used for authentication increases, and there is little contamination with environmental noise. By using the information obtained from the sound, it is possible to improve the voice authentication accuracy.

  In addition, according to the modification of the embodiment, the high frequency range portion of the characteristic parameter of the non-contact type microphone and the low frequency range portion of the contact type microphone are integrated, so that the excellent acoustic characteristics of both microphones are effective. This makes it possible to collect voices with a high S / N ratio and further improve voice authentication accuracy.

  The correspondence between the constituent elements of the claims and the constituent elements of the embodiment is as follows. That is, the non-contact type microphone 1 constitutes the first microphone, the contact type microphone 2 constitutes the second microphone, the arithmetic device 5 constitutes the feature extraction means, the comparison / collation means, and the integration means, and the storage device 5 constitutes the feature storage means. . The above description is merely an example, and when interpreting the invention, the correspondence between the items described in the above embodiment and the items described in the claims is not limited or restricted.

It is a figure which shows the structure of one embodiment. It is a figure which shows an example of the characteristic parameter of speech sound. It is a flowchart which shows the audio | voice authentication operation | movement of one Embodiment. It is a figure which shows the audio | voice authentication principle of one Embodiment. It is a figure for demonstrating a speaker authentication principle. It is a figure explaining the method to integrate and authenticate the characteristic parameter of a non-contact type microphone, and the characteristic parameter of a contact type microphone.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Non-contact-type microphone 2 Contact-type microphone 3 Microphone amplifier 4 AD converter 5 Arithmetic device 6 Storage device

Claims (5)

A first microphone that collects air conduction sound of human speech;
A second microphone that collects the body conduction sound of human speech,
Feature extraction means for extracting feature parameters from the speech of a person collected by the first microphone and the second microphone;
Feature storage means for collecting human speech using the first microphone and the second microphone, and extracting and storing feature parameters by the feature extraction means;
A comparison collation in which feature parameters are extracted by the feature extraction unit from speech uttered by a person collected by the first microphone and the second microphone, and the feature parameter stored in the feature storage unit is compared with each microphone. Means and
A voice authentication apparatus that authenticates whether or not the user is qualified based on a result of the comparison and collation by the comparison and collation means. The voice authentication device according to claim 1,
A voice authentication device that authenticates a user as a user when both of the first microphone and the second microphone obtain matching results by the comparison and verification unit. The voice authentication device according to claim 1,
A voice authentication device that authenticates a user as a user when a matching result is obtained in either one of the first microphone and the second microphone by the comparison and collation means. A first microphone that collects air conduction sound of human speech;
A second microphone that collects the body conduction sound of human speech,
Feature extraction means for extracting feature parameters from the speech of a person collected by the first microphone and the second microphone;
Integration means for generating a characteristic parameter by integrating a part of the characteristic parameter of the first microphone and a part of the characteristic parameter of the second microphone;
In advance, human speech is collected by the first microphone and the second microphone, the feature parameters of both microphones are extracted by the feature extraction means, and the feature parameters of both microphones are integrated and stored by the integration means. A feature storage means,
The feature extraction unit extracts feature parameters of both microphones from the speech of a person collected by the first microphone and the second microphone, and the feature storage unit integrates the feature parameters of both microphones. A comparison / matching means for comparing and matching with the feature parameter of the integrated result stored in
A voice authentication device that authenticates a user as a user when a matching result is obtained by the comparison and matching unit. The voice authentication device according to claim 4,
The voice authentication apparatus, wherein the integration unit integrates a high-frequency part of a characteristic parameter of the first microphone and a low-frequency part of the second microphone.

Images