JP4351659B2 - Voiceprint password key system - Google Patents

Description

  The present invention relates to a voice print password key system, and more particularly to a voice print password key system used for electronic materials (files or folders). The present invention relates to a voiceprint password key system that can enhance the security protection of electronic materials.

  At present, identifying personal identity using personal biological features is gradually becoming a major technology for personal identification. In the identification technology, biological features that can be identified at present include face identification, fingerprint identification, palm print identification, eyeball iris identification, deoxyribonucleic acid (DNA) array pair identification, and the like. In general, a personal password is used to protect the security of electronic materials. However, a simple personal password cannot provide effective security protection. There are successful decoding. Therefore, in order to increase the safety of electronic materials, other measures for protection must be sought in the security protection of electronic materials. If the cost of the identification technology structure and the actual operation are taken into account, it can be found that the identification by voiceprint is relatively suitable for the purpose of personal identification.

As a conventional voiceprint authentication system, it is determined whether or not the user has been authorized by discriminating the identity of the user by using a spectrum of voices that are unique to different users. The above system includes the following steps: a) After the user utters a voice, the end point of the voice is detected. b) Extract voice features from the voice spectrum. c) The necessity of training is determined. If “Yes”, the voice feature is used as a reference sample, and at the same time, a boundary is set. If “No”, the process proceeds to the next stage. d) Compare the design of the audio features and the reference sample. e) The distance between the two is calculated based on the comparison result. f) Compare the result of the calculation with the setting boundary. g) According to the comparison result, it is determined whether or not the user is an authorized user (see, for example, Patent Document 1).
Republic of China notice number 490655

  In the conventional voiceprint authentication system as described above, the system is used for a mobile phone, and a user is discriminated by extracting unique information of voice by using a voice spectrum analysis method. Is the method. In the conventional voice authentication system, the main numerical value of each time frame (time frame) is mainly used for comparison with the boundary line set by the user, the voice start point and end point are determined, and then the Prince- By converting the detected audio signal using the Bradley filter, the corresponding audio spectrum design is obtained. The voiceprint of the user is determined by comparing the design of the speech spectrum with a reference speech spectrum sample stored in advance. In short, in a conventional voiceprint authentication system, the design must be matched and the distance calculated, and if the calculated distance does not exceed the boundary line, the user passes the voiceprint discrimination. Can do. However, the distance between the reference sample and the test sample must be calculated when performing the above-described design matching and distance calculation. Actually, the database space occupied by the reference sample is considerably large, so that a relatively large database space is required and a longer file transmission time is required. In addition, once a reference sample is too large for the safety protection of personal electronic materials, there is a problem in that it is not applicable to loading a storage medium with a limited storage space. Thus, by further improving the problem that the reference sample in the conventional voiceprint authentication system as described above occupies a large space, it is possible to save the space of the database storing the reference sample, The present invention can be applied to a storage medium or device having a limited storage space.

  The present invention was invented in view of such problems, and the object of the present invention is to adopt a voiceprint authentication system and perform front-end processing before extracting features of speech. We will provide a voiceprint password key system that reduces the amount of processing material and increases the probability of discrimination by extracting valid voice information from the original input voice data and filtering and removing invalid voice information. It is what.

  In addition, in the present invention, a voiceprint key (voiceprint) is generated using a voiceprint authentication system, for example, one voiceprint feature value is selected, and the electronic material is encrypted or decrypted using the voiceprint key. It is intended to provide a voiceprint password key system that can be used to protect the electronic material by constructing a voiceprint password key by performing a decryption operation.

  A first object of the present invention is to construct a voiceprint password key by performing an operation of encrypting or decrypting electronic material using a voiceprint key, so that a voiceprint password key that can enhance the safety of the electronic material. It is intended to provide a system.

  The second object of the present invention is to adopt a voiceprint authentication system, extract valid voice information from the original input voice data using front end processing, and filter invalid voice information before training and testing of voice feature extraction. Therefore, it is an object of the present invention to provide a voiceprint password key system that can reduce the processing amount of materials and increase the probability of discrimination by removing them.

  The third object of the present invention is to reduce the effective speech information by using the front end processing in the test, extract the speech feature, and obtain the most similar path by using the Viterbi arithmetic method. Allows you to calculate and store model parameters (desired values and number of mutations for each state), and also to calculate model parameters and test similar probabilities between speech features during speech training or testing It is intended to provide a voiceprint password key system that can simplify training and testing operations because voiceprint feature values can be acquired by itself.

In order to achieve the above object, a voiceprint password key system according to the present invention is as follows. That is,
The voice print authentication system (1) and the voice print key (Kc) are used. The voiceprint authentication system (1) is provided with a feature extraction unit. The feature extraction unit acquires a speech feature by adopting a linear prediction cepstrum coefficient and a mel frequency cepstrum coefficient for the effective speech information obtained from the original input speech material by the front end processing unit. The voiceprint authentication system (1) obtains a voiceprint feature file (31) by obtaining a route most similar to the voice feature using the Viterbi calculation method from the voice feature. The voiceprint feature file (31) is used to generate a voiceprint feature value, and the voiceprint key (Kc) is selected from the voiceprint feature value, and the electronic document file (32) is encrypted. It is used to perform decoding operations. When the error between the voiceprint feature file (31) and the input voice is lower than a predetermined value, it is considered that the voice test work has been passed and the electronic material file (32) is permitted to be opened. . Further, when the error becomes higher than a predetermined value, it is regarded that the voice test work is not passed, and the electronic material file (32) is not permitted to be opened.

The voiceprint key (Kc) may be a fixed voiceprint password key applied to a personal computer, notebook computer, personal digital assistance, or mobile phone. The voiceprint feature file (31) is provided with a general key (K) for encrypting and decrypting the voiceprint feature file (31), and the voiceprint feature file (31) is stored in an encrypted state by the key (K). You may make it do. The voiceprint key (Kc) may be a portable voiceprint password key applied to network or recording medium transmission. The voiceprint feature file (41) includes a general key (K) for encrypting and decrypting the voiceprint feature file (41), and the voiceprint feature file (41) is stored in an encrypted state by the key (K). The electronic material file (42) is stored in an encrypted state by the portable voiceprint password key, and the encrypted voiceprint feature file (41) and the encrypted electronic material are stored. The file (42) may be a serial file (40) connected in series. Further, an information authentication code is generated in the serial file (40) of the encrypted voice print feature file (41) and the encrypted electronic material file (42) by using the information authentication code generation structure. When decrypting, the authentication code may be compared with the information authentication code, the voiceprint feature file (41), and the electronic material file (42). Further, the information authentication code is encrypted by performing an operation of encrypting the information authentication code of the serial file (40) using the key (K), and the key (K) further stores the information authentication code. It may be used to perform a decoding operation. The voiceprint authentication system (1) includes a storage unit and a calculation unit in addition to the front end processing unit and the feature extraction unit, and the front end processing unit includes the voiceprint authentication system (1). Is used for front end processing of the original input audio material, and after completing the classification of the effective audio information and the invalid audio information, the effective audio information is extracted again. The voice feature may be used to store the voice feature, and the calculation unit may be used to calculate the stored voice feature and the input voice feature. . The voiceprint authentication system (1) includes a training system (10) that is executed when an ID number for confirming the identity of the user is not registered. The training system (10) By using the front end processing unit and the feature extraction unit, the voice feature may be acquired from the original input voice material. Also, the training system (10), by using the Viterbi calculation method to another, by acquiring the most similar path to the audio features from the speech features, the model parameters as the voiceprint characteristic file (31) May be calculated and stored. The voiceprint authentication system (1) includes a test system (20) that is executed when an ID number for confirming the identity of the user has been registered. The test system (20) By using the front end processing unit and the voiceprint authentication system (1), a voice feature may be obtained from the original input voice material.

According to the voiceprint password key system of the present invention, the voiceprint password key is constructed by performing the operation of encrypting or decrypting the electronic document file by using the voiceprint key, thereby enhancing the safety of the electronic document file. There is an advantage that can be.

  According to the voiceprint password key system of the present invention, the voiceprint authentication system is adopted, and before the training and testing of the voice feature extraction, the valid voice information is extracted from the original input voice data using the front end processing, and the invalid voice is recorded. By filtering out information, there is an advantage that the processing amount of the material can be reduced and the probability of discrimination can be increased.

  According to the voice print password key system of the present invention, the effective voice information is reduced by using the front end processing in the test, the voice feature is extracted, and the most similar path is obtained by using the Viterbi arithmetic method. By acquiring, model parameters (desired values and number of mutations in each state) can be calculated and stored, and also during speech training or testing, model parameters are calculated and the probability of similarity between speech features Since the voiceprint feature value can be obtained only by testing the training, there is an advantage that the work of training and testing can be simplified.

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

  The voiceprint password key system according to the embodiment of the present invention employs a voiceprint authentication system to perform training or test work on the original input voice material. FIG. 1 is a block diagram of the flow of the voiceprint authentication system employed by the voiceprint password key system.

  Referring to FIG. 1, a voiceprint authentication system 1 employed by a voiceprint password key system according to an embodiment of the present invention includes a training system 10 and a test system 20, so that training or testing of a source input voice material is performed. Do work. In addition, the voiceprint authentication system 1 includes a front end processing unit, a feature sampling unit, a storage unit, and a calculation unit. The front end processing unit and the feature extraction unit are used by the training system 10 and the test system 20 to perform front end processing and feature extraction, the storage unit is used to store the audio features, and the arithmetic unit is used to store the stored audio features. Used to compute input speech features.

  When a user intends to enter the voiceprint authentication system 1 of the present invention, first, an ID number is input to confirm the status. Then, the voiceprint authentication system 1 searches the database for whether or not the input ID number has already been registered based on the input ID number. If the input ID number is unregistered, the voice material of the input ID number is registered and stored by entering the training system 10 and performing voice training work. If the input ID number has already been registered, the voice feature of the input ID number is already stored by entering the test system 20 and performing a voice test operation. It is determined whether or not the audio material matches.

その中、

は原始信号で複数個のＤ次元の音フレームに分け、

は所属する確率であり、

は背景背の雑信号の要望値であり、Σ_i は背景の雑信号の変異数である。ここで、

の中のD=256は一つの定数で、故に省略して計算に入れず、方程式(1)を下記の如く簡単にする。

上記方程式の指数の運算は運算のデータを運算する上では過大になるかもしれず、故にそれを対数に取った後、方程式(2)を下記の如く簡単にする。

原始入力音声資料前端256点を抜取り、短間隔エネルギーとゼロ交差率の要望値と変異数を計算し、それから二個の数値と原始入力音声資料を方程式(3)に代入して運算を行う。短間隔エネルギーとゼロ交差率の分布確率を利用して有効音声情報と無効音声情報を区分し、無効音声情報を濾過して除去することにより、資料の量を減らすだけではなく、有効音声情報を正確に抜取ることができる。 Before extracting the features of the sound, the effective sound information is extracted from the original input sound material by using the front end processing unit, and the invalid sound information is filtered and removed. In the present invention, detection including a short interval energy (Short-Energy) and a zero-crossing rate is performed. In the present invention, a calculation method combining the Gaussian distribution probabilities is adopted, and the equation is as follows.

Among them,

Is a primitive signal divided into multiple D-dimensional sound frames,

Is the probability of belonging,

Is the desired value of the background miscellaneous signal, and Σ _i is the number of variations of the background miscellaneous signal. here,

D = 256 in is a constant, so it is omitted and not included in the calculation, and equation (1) is simplified as follows.

The exponential operation of the above equation may be over-running the operation data, so after taking it logarithm, equation (2) is simplified as follows.

Extract 256 points from the front edge of the original input audio material, calculate the desired value and the number of mutations of short interval energy and zero crossing rate, and then substitute the two numerical values and the original input audio material into equation (3) for the calculation. By separating effective speech information and invalid speech information using short interval energy and zero-crossing rate distribution probability, and filtering and removing invalid speech information, not only reducing the amount of data but also valid speech information Can be extracted accurately.

Ｋは音フレームの数を考慮することができる。
一階ケプストラムモデルパラメーターの公式(4)が複雑過ぎて、故にそれを簡単にし、下記の各式は前後各二個のタイムフレームのみを考慮した時、方程式は下記の如く簡単化される。

方程式(5)から(9)の中、Cnはn階の特徴値であり、Lは信号中のタイムフレームの総数であり、iはタイムフレームの番号である。 In the present invention, the feature extraction unit employs model parameters of two speech discrimination features in the present invention. A linear prediction cepstrum coefficient (LPCC) and a mel frequency cepstrum (Mel Frequency) are used. Cepstrum Coefficient (MFCC). Both include 12 cepstrum model parameters (cepstral coefficients) and 12 first-order cepstrum model parameters (delta-cepstral coefficients). Cepstral model parameter c _n is partially differentiated with respect to time.

K can take into account the number of sound frames.
When the first-order cepstrum model parameter formula (4) is too complex and therefore simplified, the equations are simplified as follows, considering only two time frames before and after:

In equations (5) to (9), Cn is an nth-order feature value, L is the total number of time frames in the signal, and i is the time frame number.

  FIG. 2 shows an explanatory diagram of the relationship between the state and the sound frame by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. In training speech, a so-called “state” concept is formed in speech, and the state is a change in mouth shape and vocal tract during pronunciation. Generally speaking, every time you speak, the shape of your mouth always changes, so every state is a manifestation of a single voice change feature. Sometimes a single note warps and many states are included. One state does not have a fixed size like a sound frame, and usually one state includes several or tens of sound frames.

Referring to FIG. 2, the first state includes three sound frames, the second state includes six sound frames, and the third state includes four sound frames. In the present invention, an assumption of the relationship between the state and the sound frame is first made using a uniform cut method. Then, in the present invention, the starting model parameters (including the desired value and the number of mutations) are calculated. A new cut point is obtained by newly distributing the relationship between the sound frame and the state using the model parameter of the start. Then, a new sound frame corresponding to each state is calculated using the new cut point. Then, until the probability of similarity to the maximum is not to rise again, new distributing the relationship between the repetition sound frame and status, further each state newly calculate the corresponding sound frames.

  FIG. 3 is an explanatory diagram of a distribution model for starting sound frames and states by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. The starting distribution model performs an operation of equally dividing three audio samples as an example.

  In the starting model, the operation of dividing the sound equally is performed, and after dividing it evenly, it may not be divisible into an integer, and the remaining sound frames are divided equally into the first and last states. Referring again to FIG. 3, in the distribution model, three factors must be considered in order to evenly divide the audio samples: a) The first sound frame always belongs to the first state. b) The last one sound frame always belongs to the last one state. c) The change in the state of the sound frame is only unchanged or changes to the next. Using the Gaussian distribution probability, the probability that each sound frame belongs to each state is calculated, and the most similar path is obtained using the Viterbi arithmetic method.

  FIG. 4 shows an explanatory diagram of state change by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. Referring to FIG. 4, in three states, directions of possible state conversion of L sound frames are shown. A sound frame with a cross is considered to belong to the impossible state, and the direction of the arrow is considered to be a change path of the possible state.

  FIG. 5 is an explanatory diagram of the most similar path by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. Referring to FIG. 5, the most similar path of the sampling feature includes the first to third sound frames in the first state, the fourth to sixth sound frames in the second state, and further includes the third sound frame. The state includes the seventh to tenth sound frames.

  FIG. 6 is an explanatory diagram for equally dividing sound frames by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. Referring to FIG. 6, the three speech samples adopt a distribution after the start model in three states is equally divided. After each sound sample of the first sound sample equally divides three sound frames, the remaining two sound frames are distributed to a first state and a second state, respectively. Each audio sample of the second audio sample equally divides four sound frames. After each sound sample of the third sound sample equally divides the three sound frames, the remaining sound frame is distributed to the first state. After calculation, its maximum similarity probability is 2157.

  FIG. 7 is an explanatory diagram for firstly distributing sound frames by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. Referring to FIG. 7, after first distributing a new sound frame, the maximum similarity probability increases to 3171.

  FIG. 8 is an explanatory diagram for distributing a second sound frame by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. Referring to FIG. 8, after the second new sound frame is distributed, the maximum similarity probability increases to 3571.

  FIG. 9 is an explanatory diagram for distributing the best sound frame by the voiceprint authentication system employed by the voiceprint password key system according to the embodiment of the present invention. Referring to FIG. 9, after the sound frame is newly distributed several times, the maximum similarity probability 3571 does not rise again, so it is regarded as the best sound frame distribution. The desired value of each state and the number of mutations are calculated as model parameters. The model parameters can form a voiceprint feature file, and the voiceprint feature file can be stored in a database.

Referring to FIG. 1 again, when entering the training system 10 and performing speech training, effective training speech features can be obtained by computing equations (1) to (9). Then, the most similar path is obtained using the Viterbi arithmetic method. Then, the task of voice training can be completed by calculating the desired value and the number of mutations in each state and using them as model parameters. In speech training, the probability of similarity to the maximum is less than the scheduled reference value, the inability to pass the voice training, to finish the work of training, it must operate a new voiceprint authentication system 1. On the other hand, when the probability of similarity to the maximum becomes larger than the planned reference value, the model parameter is set as a voice print feature file by passing the voice training, and the voice print feature file is set using a general key. Encryption can be performed and stored in the voiceprint authentication system 1.

  Referring to FIG. 1 again, when the test system 20 is entered and the voice test operation is performed in the same manner, the effective test voice feature can be obtained by calculating equations (1) to (9). Further, by performing the decryption of the voiceprint feature file, the voiceprint authentication system 1 can perform the voiceprint authentication work.

Referring again to FIG. 1, the identification result is then output by computing the similarity probability between the valid test speech feature and the model parameter. In the voice identification, when the minimum similarity probability becomes larger than the scheduled reference value, the voiceprint authentication system 1 can be entered by passing the voice identification. On the other hand, if the minimum similarity probability is smaller than the planned reference value, the voice identification cannot be passed and the test work is completed. Disappear.

FIG. 10 is a block diagram showing a flow in which the voiceprint password key system according to the first embodiment of the present invention performs encryption and decryption using a fixed voiceprint password key. Referring to FIG. 10, the voiceprint password key system 3 according to the first embodiment of the present invention includes a voiceprint key Kc . The voiceprint key Kc is used to calculate the encryption or decryption of the electronic material file 32 , thereby forming a fixed voiceprint password key. The fixed voiceprint password key is applied to a personal computer (PC), a notebook computer (NB), a personal digital assistance (PDA), a mobile phone, or the like.

1 and 10 again, the voiceprint password key system 3 of the present invention employs the training system 10 of the voiceprint authentication system 1. The voiceprint authentication system 1 is used to generate a voiceprint feature value, and selects the voiceprint key Kc from the voiceprint feature value. The training system 10 can provide a voiceprint feature file 31 and preferably extracts 32 bits from the voiceprint feature file 31. The so-called voiceprint key Kc is the storage end and the extraction end of the electronic material file 32 , and all the same voiceprint features must be input. When the electronic material file 32 is stored, the voiceprint key Kc can be acquired in advance by using the training work of the training system 10 of the voiceprint authentication system 1. At this time, the voiceprint key Kc is used to calculate the encryption of the electronic material file 32. Further, by storing the encrypted electronic material file 32 in a specific position, the encryption of the electronic material file 32 is performed. Complete the work. In the present invention, the encryption operation is preferably performed by employing an advanced encryption standard (AES) and a symmetric encryption method.

Referring to FIGS. 1 and 10 again, when extracting the electronic material file 32 , the voice print feature file 31 is first decrypted using a general key K. Then, voice test work is performed using the test system 20 of the voiceprint authentication system 1 to compare the voiceprint feature file 31 with the input voice. If the error between the voiceprint feature file 31 and the input voice sound is lower than the expected value, it is considered that the voice test work has passed, and once the voice test work is passed, the electronic material file 32 can be permitted to open. . On the other hand, if the error between the voiceprint feature file 31 and the input voice sound is higher than the expected value, it is considered that the voice test work has not been passed, and once the voice test work is not passed , the electronic material file 32 is opened. It is configured not to allow

FIG. 11 is a block diagram showing a flow in which the voiceprint password key system according to the second embodiment of the present invention performs encryption and decryption using a portable voiceprint password key. Referring to FIG. 11, the system 4 of the voiceprint of Example 2 of the present invention that have a voiceprint key Kc. The voiceprint key Kc is used to calculate the encryption or decryption of the electronic material file 42 , thereby forming a portable voiceprint password key. The portable voice print password key is applied to a compact disk (CD), a flexible disk (Flexible Disk), a flash disk (Flash Disk), an MO disk (MO Disk), or network transmission.

1 and 11, the voiceprint password key system 4 of the present invention employs the training system 10 of the voiceprint authentication system 1, and the training system 10 can provide a voiceprint feature file 41. The voiceprint authentication system 1 is used to generate a voiceprint feature value, and selects the voiceprint key Kc from the voiceprint feature value. When the electronic material file 42 is stored, the voiceprint key Kc can be acquired in advance by using the training work of the training system 10 of the voiceprint authentication system 1. At this time, the voiceprint feature file 41 is inserted into the electronic material file 42, that is, the voiceprint feature file 41 occupies only 2K to 6K bytes of the electronic material file 42. The encrypted electronic material file 42 is generated by performing encryption calculation of the electronic material file 42 using the voiceprint key Kc. Further, by performing encryption operation on the voiceprint feature file 41 using a general key K, an encrypted voiceprint feature file 41 is generated. Further, the serial file 40 is obtained by connecting the encrypted voice print feature file 41 and the encrypted electronic material file 42 in series. In addition, an information authentication code is generated from the serial file 40 of the encrypted voice print feature file 41 and the encrypted electronic material file 42 through an information authentication code generation structure. In the present invention, preferably, an information authentication code is generated by employing a secure summarization algorithm (SHA) calculation method, and the calculation of encryption of the information authentication code is performed using a general key K. By doing so, an encrypted information authentication code is generated. The encrypted electronic material file 42, voiceprint feature file 41 and information authentication code can form a portable file, that is, the electronic material file 42 is encrypted and transmitted via a network or a recording medium. can do.

Referring again to FIGS. 1 and 11, when the electronic material file 42 is extracted in the computer system, the encrypted voiceprint feature file 41 and the encrypted information authentication code are first decrypted using a general key K. Is obtained, the voiceprint feature file 41 and the information authentication code are obtained. Then, voice test work is performed using the test system 20 of the voiceprint authentication system 1 to compare the voiceprint feature file 41 with the input voice. If the error between the voice print feature file 41 and the input voice sound is lower than the expected value, it is considered that the voice test work has passed, and once the voice test work is passed, the electronic data file 42 is allowed to be opened by the voice print key Kc. can do. On the other hand, if the error between the voiceprint feature file 41 and the input voice sound is higher than the scheduled value, it is considered that the voice test work has not been passed, and once the voice test work is not passed , the electronic material file 42 is opened. It is configured not to allow

  Referring to FIG. 11 again, the voice print feature file 41 and the electronic material file 42 are finally compared with the information authentication code. If the voiceprint feature file 41 and the electronic material file 42 match the information authentication code, the electronic material file 42 can be represented in the computer system. On the other hand, if the voice print feature file 41 and the electronic data file 42 do not match the information authentication code, the electronic data file 42 cannot be represented in the computer system and must be separated from the voice print password key system 4. Don't be.

As described above, according to the conventional voiceprint authentication system, the database space occupied by the reference sample is quite large, so a relatively large database space is required and a longer file transmission time is required. Although there was a problem, according to the voiceprint password key system of the present invention shown in FIG. 1, before extracting the features of the voice, the valid voice information is extracted from the original input voice data by using the front end processing and invalidated. By filtering and removing the voice information, it is possible to reduce the processing amount of the material and increase the probability of discrimination. In addition, a voiceprint key is generated by generating a voiceprint key using a voiceprint authentication system, selecting one voiceprint feature value, and performing an operation of encrypting or decrypting an electronic material file using the voiceprint key. Can be used to protect electronic material files .

  The present invention may be implemented in other ways without departing from the spirit and essential characteristics thereof. Accordingly, the preferred embodiments described herein are illustrative and not limiting.

It is a flowchart which shows the flow by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the relationship between the state by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts, and a sound frame. It is a figure which shows the distribution model of the start of a sound frame and a state by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the state change by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the path | route most similar by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the state which divides | segments the sound frame equally by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the state which distributes a sound frame newly by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the state which distributes the sound frame newly by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a figure which shows the state which distributes the best sound frame by the voiceprint authentication system which the system of the voiceprint password key of the Example of this invention employ | adopts. It is a block diagram which shows the flow which the system of the voiceprint password key of Example 1 of this invention performs encryption and decoding using a fixed voiceprint password key. It is a block diagram which shows the flow which the system of the voiceprint password key of Example 2 of this invention performs encryption and decoding using a portable voiceprint password key.

DESCRIPTION OF SYMBOLS 1 Voiceprint authentication system 10 Training system 20 Test system 3 Voiceprint password key system 31 Voiceprint feature file 32 Electronic material file 4 Voiceprint password key system
40 Serial file 41 Voiceprint feature file 42 Electronic material file
K key
Kc Voiceprint key

Claims (11)

A voiceprint password key system using a voiceprint authentication system (1) and a voiceprint key (Kc),
The voiceprint authentication system (1) is provided with a feature extraction unit,
The feature sampling unit is adapted to acquire a speech feature by adopting a linear prediction cepstrum coefficient and a mel frequency cepstrum coefficient with respect to the effective speech information obtained by the front end processing unit from the original input speech material,
The voiceprint authentication system (1) obtains a voiceprint feature file (31) by obtaining a route most similar to the voice feature using the Viterbi calculation method from the voice feature.
The voiceprint feature file (31) is used to generate voiceprint feature values,
The voiceprint key (Kc) is selected from the voiceprint feature values and is used to perform operations for encryption and decryption of the electronic material file (32).
If the error between the voiceprint feature file (31) and the input voice is lower than the expected value, it is considered that the voice test work has passed and the electronic document file (32) is allowed to open, and the error is scheduled. A voiceprint password key system configured so as not to allow the electronic material file (32) to be opened if it exceeds the value and is regarded as not passing through the voice test operation. The voiceprint key (Kc) may be a personal computer, notebook computer, voiceprint of claim 1, wherein that you have a voiceprint stationary to be applied to a personal digital assistant or mobile phone System. Ri Contact comprise common key (K) to encrypt and decrypt the voiceprint characteristic file (31),
The voiceprint characteristic file (31), the system of the voiceprint of claim 2, wherein that you have come to be stored in a state of being encrypted by the key (K). The system of the voiceprint key (Kc) is voiceprint of claim 1, wherein that you have a voiceprint portable applied to a transmission network or a recording medium. Includes a common key (K) to encrypt and decrypt the voiceprint characteristic file (41),
The voiceprint feature file (41) is stored in an encrypted state by the key (K), and the electronic material file (42) is stored in an encrypted state by the portable voiceprint password key. And
Encrypted the voiceprint characteristic file (41) and encrypted the electronic document file (42), voiceprint of claim 4, wherein that you have a connected series files in series (40) Password key system. Using the information authentication code generating structure, to generate information authentication code in the series file (40) encrypted the voiceprint characteristic file (41) and encrypted the electronic document file (42), decoding when performing reduction, the information authentication code and the voiceprint characteristic file (41) and voiceprint password according to claim 5, characterized in that is adapted to perform comparison of the authentication code to said electronic information file (42) Key system. The information authentication code is encrypted by performing an operation of encrypting the information authentication code of the serial file (40) using the key (K) , and the key (K) decrypts the information authentication code. The voiceprint password key system according to claim 6, wherein the voiceprint password key system is used to perform the following calculation. The voiceprint authentication system (1) is adapted to the configuration with the front section and the storage section and mathematical operation unit in addition to the feature sampling unit,
The front end processing section, by used to front end processing the raw voice article in the voiceprint authentication system (1), after completing the division of the effective audio information and disabling the audio information, the valid voice information again Ri you become so that sampling,
The storage unit is adapted to so that used to store the speech feature,
The mathematical operation unit, the voiceprint of claim 1, wherein the speech feature stored, that is adapted to be used to mathematical operation input speech feature system. The voiceprint authentication system (1) includes a training system (10) that is executed when an ID number for confirming the identity of the user is not registered .
The training system (10), said by using the front section and the characteristic sampling unit, voiceprint password according to claim 1, characterized in that is adapted to acquire the speech feature from the raw voice article Key system. The training system (10), by using the Viterbi calculation method to another, by acquiring the most similar path to the audio features from the audio feature, calculates the model parameters as the voiceprint characteristic file (31) 10. The voiceprint password key system according to claim 9, wherein the voiceprint password key system is stored. The voiceprint authentication system (1) includes a test system (20) that is executed when an ID number for confirming the identity of the user has been registered.
Said test system (20), by the use of the front section and the voiceprint authentication system (1), according to claim 8, wherein that you have come to acquire speech features from the raw voice article Voiceprint password key system.

Images