TWI234762B - Voiceprint identification system for e-commerce - Google Patents

Abstract

An e-commerce method includes steps of: accessing user's login through an electronic communication means; using an identification device to recognize a password; using the identification device to verify whether the user's registration for voiceprint identification; using a voiceprint identification system to identify or register for voiceprint identification; and using an identification device to decide for allowing or rejecting the user for proceeding e-commerce.

Description

1234762

[Technical field to which the invention belongs] The present invention relates to a voiceprint verification system for e-commerce transactions. In addition to voiceprint verification for e-commerce transactions, choose a language, eight cloth probability, dynamic time calibration algorithm, and hiding. The Markov Pyramid II uses a Viterbi algorithm to obtain a voiceprint verification system that uses the most similar path to model parameters. " " [Prior art]

Conventional e-commerce transaction methods, such as the Republic of China Patent Bulletin No. 38541 6 "E-commerce system" invention patent, which reveals an e-commerce that provides archiving safety for a transaction record (transaction 10g) on a network A system [c〇mmerce system], which includes:-a session key generator [session and creator] to generate a session key to encrypt the transaction record; an easy record encryptor [encryptor] for encryption use The session key | the transaction record; and a transaction record sender for sending the encrypted transaction record to an archive server on the network. However, the 3rd, 3rd, 5th, 4th, and 6th encryption only encrypted the parent record for data transmission and storage, and did not identify the user. Another conventional e-commerce transaction method, such as the Republic of China Patent Bulletin No. 550477 ^ website account method, system and computer-readable media and e-commerce management from a central location "invention patent, which discloses a method for managing a user's A method for conducting online [central website] financial transactions on a destination e-commerce website, which includes: logging in users to a destination e-commerce platform; and, weekly stations, generating the sole use of the user on the central website By

Page 71234762 V. Description of the invention (2) Name and password; use the unique user name and password to register on one or more destination websites; transmit an activation instruction to a processing finance 易易 2 2 Ϊ Ϊ Used to activate a credit or debit card account for the user. When the #used or debit card account is enabled, send one of the credit or debit card account via the local e-commerce website = and transmit — Revocation order to the financial institution to withdraw: the "used or debit card account; when the credit or debit card account is activated, the financial institution only accepts and processes requests from the e-commerce website; Among them, when the credit or debit card account was revoked, the financial institution refused to pay the request. However, the 5504m tiger used the user's unique name and password to identify him, so the bean and the fork code leaked. Doubt. 88:12 i: The e-commerce transactions No. 38541 6 and No. 550477 need to be improved so that users can be accurately identified. &Amp; Voice verification: law, such as the Republic of China Patent Bulletin 490655 J: Method wrong user identification information with its sound devices "designed f invention different user-specific sound wrong user identification information body injury 'to determine whether the user is authorized Cloth. The method includes the following steps: "Li, A_ Detect the end of the speech just after the user has spoken; [2], since =" = feature; [3], decide whether training is needed-Jieyang, "" and "two features as Take a reference sample and set the symbol two to go to #No ”, then proceed to the next step— [4], pattern comparison of the voice special samples; [5], according to the distance between the two; [6], will Ratio of the calculation result to the set limit

· _ \ L (X; 〇-5 \ FIVH Ci) NTINli'NTS \ pj (() 35 | Page 81234762 V. Description of the invention (3) comparison; [7], determine whether the user is based on the comparison result It is an authorized user. This method is used in mobile phones. It uses sound spectrum analysis method to extract the unique information of speech to identify the user. This 490655 mainly uses the main of each frame [fraine] The value is compared with the limit set by the user. After determining the start and end points of the speech, the detected speech signal is converted by the Princen-Bradley filter in order to obtain its corresponding sound spectrum pattern. The sound spectrum pattern and the pre-stored The reference sound spectrum samples are compared to identify the user's voiceprint. In short, the 490655 requires pattern matching and distance calculation. 'If the calculation distance does not exceed the limit, the user can pass the voiceprint. Identification. However, when performing pattern matching and distance calculations, the No. 49 0 65 5 must calculate the distance between the reference sample and the test sample. In fact, 'the reference sample occupies a considerable amount of space in the database Large, so it comes but requires a larger database space and longer file transfer time. If 4 can apply the voiceprint verification technology to e-commerce transactions, it has the disadvantage of extending the transaction time. Therefore, this section 490655 It is still necessary to further improve the problem of the space occupied by its reference samples, so as to save the database space for storing the reference samples' to avoid the limitation of the number of users. By reducing the bit method of the reference samples, the voiceprint verification can be accelerated. The time required, and the recognition rate can be improved, so that the voiceprint verification technology can be applied to e-commerce transactions, which can shorten the transaction time. In view of this, the present invention improves the above-mentioned disadvantages. Identify users with voiceprint verification system

c: \ L (K; (). 5 uq VE C () NT l Nl-NTS \ PK9 3 51. P ul Page 9 1234762 5. The invention is explained (4), and the voiceprint quasi-performance algorithm and hidden The most similar path. [Summary of the Invention] The main system of the present invention is the secondary system of the present invention, which is in addition to identifying the user. In addition to the probability of distribution, and using Viterbi numbers, the present invention makes the account of the user according to the present invention continuous. The basic knowledge of the households; profit knowledge; and the development department,-to conduct training front-end processing; then again as a model data; use a discernible voice to train the training department or the measurement department to use the calculation of this type of parameter verification system In addition, it combines the probability of Gaussian distribution, the Tibetan Markov model, and uses Viterbi ^ time calibration to calculate the model parameters. / 、 The purpose of obtaining an e-commerce transaction is to provide an e-commerce transaction shaw for e-commerce transactions, using the voiceprint verification system :: certificate system This makes the present invention have a powerful effect of improving the recognition rate. The purpose of identification is to provide a voice verification of e-commerce transactions. In addition to the verification of the voiceprint of e-commerce transactions, its dynamic time calibration algorithm and hidden type horse ;, Gao Yan's algorithm to obtain the most similar path, so that the calculation model = has the effect of simplifying training and testing operations., | Multi-service delivery: Yi Fang ☆, this method includes the steps of ordering records in the guest house, using The device verifies whether the recognizable device has applied for voiceprint = print verification system chooses to perform voiceprint recognition or register voiceprint recognition device to decide whether to allow or deny e-commerce transactions. The print verification system includes a front-end processing unit, a feature extraction Fetch, and a test system to test the original input voice data. On the training voice, the training system uses the original input voice data to extract the effective training voice feature extraction unit to extract the effective training voice feature; Efficiently train speech information to obtain the most similar path so that, also on test speech, the test system uses the previous

1234762 V. Description of the invention (5) tf Γ: Extract the valid test voice information from the original turn-by-round voice data; —Run | ^ Feature extraction section to retrieve the valid test voice feature; Similarity between speech features and model parameters in order to output [implementation] This and other objects, features, and advantages can be made more explicit, and detailed: Ϊ 举 citing the preferred embodiment of the present invention, and with the accompanying drawings of the preferred embodiment of the present invention, the voiceprint verification voice of e-commerce transactions = = Jin Jia In the embodiment, an e-commerce transaction receiving device performs login. This company :; Set: Including account 2 by 1 comDuter 1 ^ ^ s into the details [persona 1 M ^ 1 ATM [Automated Teller 广

Mach; ne], credit card swiping machine [creducar 4], two Γ can be connected for ordinary business transactions. : Texture verification center, this voiceprint verification center is optional = = Zhili-ΐίϊ: The voiceprint verification center uses an identifiable band to set the confirmation of the customer's basic information, and the identifiable device includes the editing circuit and so on. In addition, the voiceprint verification center identification logic A 夂 ZHAO Feng-TU has a voiceprint verification system. The customer said, as shown in a picture, and then, the recognizable device checks it. Please compare the voiceprint 'to generate whether the customer needs to enter it. fruit. The voiceprint verification center returns the result to the connection bucket

: \ L (Xj (). 5 \ I-IVF: C () NTINENTS \ PKy35l.ptd Page 111234762, Description of Invention (6)) for subsequent e-commerce transaction procedures. The first figure reveals that the present invention is preferred The block diagram of the process of the voiceprint verification system of the embodiment. As shown in FIG. 2 with reference to the second figure, the voiceprint verification system of the preferred embodiment of the present invention ^ training system 10 and a test system 20, so that the original input t ^ Material for training or test operations. The voiceprint verification system 1 further includes a = end = inside, a feature extraction section, a storage section and a computing section. The former jf management section and feature extraction section are provided for the Before the training system 10 and the test system 20 are performed, the processing and feature extraction are performed. The storage unit is used to store the voice features, and the operation unit calculates the stored voice features and the input voice features. The customer account number is input into the present invention. When the voiceprint verification system is 1, you can confirm the body knives. Then 'the system will query the database according to the input account, and the B input account belongs to the established account. If the input account has not been established, you must enter the training system 10 Perform voice training assignments to build fitness No input the voice data of the account. If the input account has been established, the t / y test system 20 performs a voice test operation in order to identify whether the input test feature matches the voice data of the input account. For the voice «month, please refer to the first and first pictures, and then, when the customer is not correct, then enter the request for the customer to enter the personal password. If the customer 2 asks for voiceprint than the personal password, it will enter the transaction rejection stage. With the customer After entering the incorrect password in turn, ask whether to apply for voiceprint identification registration. When you choose ^ to enter the correct personal identification registration, you will enter the stage of allowing transactions. Conversely, when = apply for voiceprint identification registration, you will enter the voiceprint verification system. The first one said that he would choose to apply for the voiceprint recognition and registration of the present invention to operate the training system 丨 〇 and the system 1 〇. The bang is as follows:

1234762

Before capturing voice features, the front-end processing unit is used to extract valid voice information from the original input voice data to filter out invalid voice information. The detection of the present invention includes short-time energy [Short-Energy] and zero-crossing rate [Zero-Crossing Rate]. The present invention uses a calculation method that combines a Gaussian probability distribution, and its equation is as follows: exP BU y (Bu _ Σ! -1 〇 一% (1) where: ^ is the original signal divided into several d-dimensional sound frames ,. 1,…, M are the probability of belonging, “7 is the expected value of background noise 4 is the number of variation of background 1 noise. Here, because ^, ^ T0 ^ D = 256 is a fixed value, so it is omitted Without calculation, simplify equation (1) as follows:

The exponential operation in the above formula, after taking the age of μ μ (2) After taking the logarithm, the equation (2) is simplified as follows: · It may be too large, so it is: 1η il / 2

1234762 V. Description of the invention (8) Secret = (4 tons 丨 4 (;-valence) (3) Retrieve the front-end 256 points of the original input voice data, calculate the short-range energy and zero-crossing rate expected value and the number of variations, The two numbers and the original input voice data are substituted into the equation (3) for calculation. The distribution probability of short-distance energy and zero-crossing rate is used to distinguish valid voice information from invalid voice information, and the invalid voice information is filtered, not only reduced The amount of data can also correctly capture effective speech information. In the feature extraction section, the present invention uses two speech recognition feature parameters, which include linear prediction cepstrum coefficients [L inear

Prediction Cepstrum Coefficient (LPCC) and Mel Frequency Standard Cepstrum Coefficient (MFCC) each have 12 cepstrai coefficients and 12 first-order cepstrai coefficients (delta-cepstrai ~ coefficients). Partial differentiation of cepstrum parameter ^ with time di Σ 7 2 (4) JU- especially κ is the number of frames to be considered.

Because the formula of the first-order cepstrum parameter is too complicated, the following formulas are added to consider only the two time frames before and after. The equation is simplified as follows: [2 * C (2,) + C (U)] / 5 (5)

C: \ L (X; (). 5 \ HIvr: C〇NTINENTS \ PK «) 351 .ptd Page 141234762 V. Description of the invention (9) ACi = [2 ^ C (3,«) + C (2 , λ)-C (0, «)] / 6 (6) AC ^ = [2 * 〇 ^ + 2,; 〇 十 C (i + l, ^)-C (i_l,»)-2 * C ( i-2, ")] / 10 (7) AC3f " 2 = [C (£ -l,«)-C (I-3, «)-23 < cC (Z; -4,«)] / 6 ( 8) △. Guang 1 = Bu C (Z-2, λ)-2 C (Z-3, ")] / 5 (9) In equations (5) to (9), Cn is the η order characteristic value, and L is the signal Total number of time frames, i is the time frame number. The third figure illustrates the relationship between the state of the voiceprint verification system and the sound frame in the preferred embodiment of the present invention. In training speech, speech has the concept of the so-called "state". State is the change of mouth shape and vocal tract during pronunciation. Generally speaking, every time the mouth shape changes, each state is a feature of a voice change. Sometimes a single tone may contain multiple states. A state does not have a fixed size like a sound box, usually a state contains several or dozens of sounds. ° Refer to the second figure, the first state contains three sound boxes, and the second bear contains six The sound frame and the third state include four sound frames. The fourth figure shows a schematic diagram of the initial allocation mode of the sound frame and status of the voiceprint verification system of the preferred embodiment of the present invention. This initial allocation pattern exemplifies three samples of $ Goy to perform an equalizing action. In the initial mode, the voice is evenly divided. After the equalization, it may not be able to be removed. The extra frame will be divided into the first and last states. Please refer to the third figure again. In the distribution mode, the sample voice must be divided equally. Two points 1. The first frame must belong to the first state; 2. The last one

1234762 V. Description of the invention (ίο) states; 3, The state of the sound box changes only in the probability of each state, and the probability is assigned to calculate the sentence of each frame. The hand and use the Viterbi algorithm to obtain the most similar path ϊf reveals that the state transition of the voiceprint verification system of the preferred embodiment of the present invention is not intended. As shown by Wuyuan, in three states, the possible states of the L sound frames are, ,,, and °. The sound frame is regarded as a state that cannot belong to, and the direction of the arrow is regarded as the path that this state can change. The second figure shows the most similar path diagram of the voiceprint verification system of the preferred embodiment of the present invention. A Please refer to the sixth figure. The most similar path of the extracted features has the first shape.

The state includes frames 1 to 3, and the 4th spleen. The e L includes frames 7 to 10. To 5 frame and ^. The seventh figure discloses a block diagram of a voiceprint verification system according to a preferred embodiment of the present invention. Daodao Please show the distribution of the three sample speeches after the initial modes t of the three states. After each sample voice of the first-sample voice has two voice frames, the remaining two voice frames are respectively arranged in the first state and state. Each sample speech of the second sample speech is divided into four frames. After each sample voice of the third sample voice is divided into three frames, the remaining one is divided into the first state. After calculation, the maximum similarity probability of 纟 is 215 7. Figure 8 shows the first time a voiceprint verification system of a preferred embodiment of the present invention

: \ L (X; ()-5 \ FIVE COST INENTS \ PK9 3 51 Page 16 1234762 5. Explanation of the invention (11) Schematic diagram of redistribution of sound frames. Refer to the eighth figure, 'Redistribute sound at the first time- The probability of similarity rises to 3171. Its maximum = the figure shows the schematic diagram of the voiceprint verification reallocation of the preferred embodiment of the present invention. <Please refer to the ninth figure once, and the similarity probability of the second reallocation rises to 3571. -Said 'The largest tenth picture thereof shows a schematic diagram of the voice-over frame of the preferred embodiment of the present invention. The best part of the system is similar to that shown in the ten diagrams:' Η Take the probability of 3571 no longer rising, so it regards the expected value of the maximum phase state and the number of mutations as the modulo :: Jiadao with "Kuang. Calculate each existing database. Wu number as a branch type parameter, the model parameters can be stored When practicing homework, Yun Zeng: J: ⑴ 在, 々 5 HAI training system 10 then use the dimension; =; expected value and the number of mutations; = = similar;! Cheng; γ γ various states 'state 2-tone training on' its maximum Similar chances for small businesses. The reference value &amp;, non-texture verification system 1; otherwise °, &quot; 畀 &quot;, so when the sound must be re-operated, the probability of similarity is greater than the predetermined reference value system 1 through voice training, and the model parameters are stored. In this voiceprint verification system, refer to the first% — ^ to enter the allowed transaction phase. When Hada 70% applies for voiceprint identification registration, that is, C: \ L (X; ()-5 \ F 丨 VE CONT 丨 NENTS \ PKy3 51 · Page 17 1 1234762 V. Description of the invention (12) Please refer to the first and second figures again, the test account 20 has been established for the turn-in account for voice test operations. The voiceprint of the present invention operates the test system The details of 2 0 are as follows: For Predicate 4, the same goes into the test system 20 for the voice test. Equations (1) to (9) are obtained to obtain valid test voice features. Please refer to the second figure again, and then perform the calculation. The probability of similarity between the test voice and the model parameters in order to output a recognition knot. ― In the five senses, when the smallest similarity probability is greater than the predetermined reference value, f is identified by the group ^ ^, and thus can leave the voiceprint verification system! Enter the follow-up electronic business The parent program, on the other hand, its minimum similarity probability is less than the predetermined reference value and cannot pass voice recognition and end the test operation. Therefore, it must leave the voiceprint inspection = system 1 and refuse to conduct subsequent e-commerce transaction procedures. As shown in the two figures, finally, the identifiable device decides whether to allow or deny e-commerce transactions based on the test results of the test system 20 of the chain verification system 1. Although the present invention has been disclosed in the foregoing preferred embodiment, it is not Limitation = the present invention. Any person skilled in this art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention is defined by the scope of the attached patent application. Prevail. : \ L (X; 〇-5 \ !; IVI: C〇NTINENTS \ PKy351 .ptd Page 18 1234762 Brief Description [Schematic Illustration] Figure 1: Voice of E-commerce Transactions in the Preferred Embodiment of the Present Invention Flowchart of the voiceprint verification system. Figure 2: Block diagram of the voiceprint verification system of the preferred embodiment of the present invention. Figure 3: Schematic diagram of the state of the voiceprint verification system and the relationship between the sound boxes of the preferred embodiment of the present invention Figure 4: Schematic diagram of the initial allocation mode of the voice frame and status of the voiceprint verification system of the preferred embodiment of the present invention. Figure 5: State transition of the voiceprint verification system of the preferred embodiment of the present invention. Figure 6 : Schematic diagram of the most similar path of the voiceprint verification system of the preferred embodiment of the present invention. Turtle FIG. 7: The equalization diagram of the voiceprint verification system of the preferred embodiment of the present invention. FIG. 8: of the preferred embodiment of the present invention Schematic diagram of the first reassignment of the voice frame of the voiceprint verification system. Figure 9: Schematic diagram of the second reassignment of the voiceprint of the voiceprint verification system of the preferred embodiment of the present invention. Figure 10: The preferred embodiment of the present invention Best Voiceprint Verification System Schematic diagram of the voice distribution box. Drawing number description: 1 Voiceprint verification system 10 Training system 20 Test system

C: \ L (X; 〇-5 \ I; IVi: CONTINUNTS \ PK *) 351 .pul page 19

Claims (1)

1234762 6. Scope of patent application, an electronic customer account using one that can be identified and one that should be recognized; verifying the original input invalid voice at a front end according to the application-specific voiceprint; a feature extraction feature; a business transaction method, which Including steps: login by a connected device; identifying the device to confirm the customer's basic information; checking whether the device has applied for voiceprint comparison; voiceprint identification system for voiceprint identification; and the device determining to allow or deny e-commerce transactions. Lee range first! The item's e-commerce transaction method system includes: Qianghaili, which is used for front-end processing. The voiceprint verification system = voice data, and thus completes the distinction of valid voice information and information, and then retrieves valid voice information; It is used for retrieving the valid voice information. The May 1 storage unit stores the voice feature; 丨 transportation; the computing unit'uses the stored voice feature and input voice feature force according to the second item of the scope of the patent application In the e-commerce transaction method, 1 The voiceprint verification system further includes a training system that uses the front-end feature extraction section to obtain a model of the original input voice data ^. 2. The e-commerce transaction method according to item 4 of the scope of patent application, wherein the training system of the voiceprint verification system uses Viterbi algorithm ^ to obtain ^ similar paths in order to calculate model parameters for storage. Also, according to the e-commerce transaction method according to the scope of the patent application, the voiceprint verification system further includes a test system using the front-end processing unit and 1234762 6. Scope of patent application Feature extraction section to obtain the voice features of the original input voice data. 6. The e-commerce transaction method according to item 1 of the scope of patent application, where the voiceprint verification system enters the personal password when no voiceprint comparison is applied. 7. The e-commerce transaction method according to item 6 of the scope of patent application, When the correct personal password is entered, whether to apply for registration of voiceprint recognition phase is entered. _ : \ L (X; ()-5 \ FIVE CONTINENTS \ PK &lt;; 3 51. p t d p. 21