CN1848165A - E-Commerce Transaction Method - Google Patents

Abstract

The present invention is an e-commerce transaction method comprising the following steps: a customer account is logged in by a connection device; a recognizable device is used to confirm the customer's basic information; the recognizable device verifies whether a voiceprint comparison has been applied for; a voiceprint verification system is used to select voiceprint recognition or registered voiceprint recognition; and the recognizable device decides to allow or refuse e-commerce transactions.

Description

Electronic business transaction method

Technical field

The present invention is about a kind of electronic business transaction method, it is particularly to carry out outside the vocal print checking of e-commerce transaction, in addition in conjunction with Gaussian distribution probability, dynamic time calibration algorithm and concealed Marko husband pattern, and profit comes Viterbi (Viterbi) algorithm to obtain the most similar path, so that the vocal print verification system of computation model parameter.

Background technology

Existing electronic business transaction method, patent of invention as TaiWan, China patent announcement No. 385416 " e-commerce system ", it discloses the e-commerce system (commerce system) that a kind of transaction record (transaction log) on network provides deposit safety (archiving safety), and it comprises: a dialogue key generator (session key creator) is in order to produce a dialogue key to encrypt this transaction record; Transaction record encryption equipment (encryptor) is in order to encrypt this transaction record that uses this session key; And the transaction record transmitter is sent to deposit server (server) on this network in order to this has been encrypted transaction record.Yet, only transaction record is encrypted so that carry out data transmission and storage, not at user's identification identity for these No. 385416 in addition.

Another existing electronic business transaction method, patent of invention as TaiWan, China patent announcement No. 550477 " method of website account, system and computer fetch medium reach the managing eBusiness from middle position ", its announcement is a kind of in order to manage the method an of user (central web site) financial transaction on the enterprising line of destination e-commerce website, and it comprises: login the user to the destination e-commerce website; Produce unique user title and the password of this user in this central web site; Utilize this unique use user's title and password or website, a plurality of destination on register; Credit card or the debit card account of financial institution in order to start this user who instructs to processes financial transactions enabled in transmission one; When this credit card or debit card account are initiate mode, send this credit card or debit card account's payment request to this financial institution via this destination e-commerce website; And the transmission one cancel an order to this financial institution to cancel this credit card or debit card account; Wherein when this credit card or debit card account were initiate mode, this financial institution only accepted and handles the payment request that receives from this e-commerce website; And wherein when this credit card or debit card account when cancelling state, payment request is refused by this financial institution.Yet, only utilize user's unique name and password identification identity No. 550477, so it has the doubt of password leakage in addition.

In brief, the e-commerce transaction of No. 385416 and No. 550477.Need progressive the improvement, so that the accurate identification user's of energy identity.

As for existing vocal print verification method, patent of invention as TaiWan, China patent announcement No. 490655 " utilizing sound spectrum information to debate method and its device of knowing the user ", it utilizes the distinctive sound spectrum information identification of different users user's identity, and whether process is authorized with the decision user.This method comprises step:

(1), after the user sends voice, the terminal point of detecting voice; (2), take out phonetic feature in the sound spectrum of these voice certainly; (3), the decision whether need the training, if, then with this phonetic feature as the reference sample, set boundary simultaneously, if not, then descend step; (4), this phonetic feature and reference sample are carried out pattern relatively; (5), according to this comparative result calculating distance between the two; (6), this result of calculation and setting boundary are compared; (7), determine according to this comparative result whether this user is the authorized user.This method is used in mobile phone, and it utilizes the unique information taking-up of voice print analysis method with voice, carries out identification user's method thus.Mainly utilize for No. 490655 the main value of each time-frame (frame) and the boundary of user's setting to compare, behind the initial point and terminal point of decision voice, the voice signal that utilizes the conversion of Princen-Bradley wave filter to detect again is so that obtain its corresponding found spectrogram.This found spectrogram compares with the reference sound spectrum sample that stores in advance, with identification user's vocal print.

In brief, No. 490655 needs carry out the computing of the coupling and the distance of pattern, and when not beyond the mark as if this computing distance, the user can pass through sound-groove identification.Yet, when the computing of coupling of carrying out pattern and distance, must calculate the distance between reference sample and test sample book No. 490655.In fact, the space of the shared database of this reference sample is quite big, so it not only needs bigger database space and longer archives transmission time of needs.If when this vocal print verification technique can be applied in e-commerce transaction, has the shortcoming that prolongs exchange hour.

Therefore, still be necessary the problem that takes up room of progressive its reference sample of improvement, so can save the database space that stores reference sample, for No. 490655 to avoid the restriction of user's quantity.Utilize to reduce the method for the position of this reference sample, more can quicken vocal print checking required time, and more can promote discrimination power,, can shorten the time of concluding the business so that when the vocal print verification technique can be applied in e-commerce transaction.

In view of this, the present invention improves above-mentioned shortcoming, it is when carrying out e-commerce transaction, except utilizing the vocal print verification system to carry out identification user's the identity, and this vocal print verification system is in addition in conjunction with Gaussian distribution probability, dynamic time calibration algorithm and concealed Marko husband pattern, and utilize viterbi algorithm to obtain the most similar path, so that the computation model parameter.

Summary of the invention

Fundamental purpose of the present invention provides a kind of electronic business transaction method, and this method utilizes the vocal print verification system to carry out identification user's identity when carrying out e-commerce transaction, makes the present invention have the effect that promotes discrimination power.

Secondary objective of the present invention provides a kind of vocal print verification system of e-commerce transaction, it is except the vocal print checking of carrying out e-commerce transaction, it is in addition in conjunction with Gaussian distribution probability, dynamic time calibration algorithm and concealed Marko husband pattern, and utilize viterbi algorithm to obtain the most similar path, so that the computation model parameter makes the present invention have the effect of simplifying training and test jobs.

According to electronic business transaction method of the present invention, this method comprises step: customer account number is logined by a coupling arrangement; But utilize a device for identifying to confirm client's basic document; But this device for identifying is checked and is applied for that whether vocal print relatively; Utilize a vocal print verification system to select to carry out sound-groove identification or registration sound-groove identification; But and should the device for identifying decision allow or refuse to carry out e-commerce transaction.

Vocal print verification system of the present invention comprises front-end processing portion, feature acquisition portion, a training system and a test macro, so that original input voice data is trained or test jobs.On training utterance, this training system utilizes this front-end processing portion from the effective training utterance information of this original input voice DAQ; Utilize this feature acquisition portion to capture this effective training utterance feature again; Carry out this effective training utterance information of computing again to obtain the most similar path, so that as model parameter.On tested speech, this test macro utilizes this front-end processing portion from this original input voice DAQ Validity Test voice messaging equally; Utilize this feature acquisition portion to capture this Validity Test phonetic feature again; Carry out between this tested speech feature of computing and the model parameter similar probability again so that export an identification result.

Description of drawings

Fig. 1 is the process flow diagram of the vocal print verification system of preferred embodiment e-commerce transaction of the present invention.

Fig. 2 is the process block diagram of the vocal print verification system of preferred embodiment of the present invention.

Fig. 3 is the synoptic diagram that concerns of the state of the vocal print verification system of preferred embodiment of the present invention and sound frame.

Fig. 4 is the sound frame of the vocal print verification system of preferred embodiment of the present invention and the original allocation pattern diagram of state.

Fig. 5 is the state exchange synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

Fig. 6 is the most similar path synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

Fig. 7 is the equal partial frame synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

Fig. 8 is for redistributing sound frame synoptic diagram the first time of the vocal print verification system of preferred embodiment of the present invention.

Fig. 9 is for redistributing sound frame synoptic diagram the second time of the vocal print verification system of preferred embodiment of the present invention.

Figure 10 is the optimal allocation sound frame synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

The figure number explanation

1 vocal print verification system, 10 training systems, 20 test macros

Embodiment

Fig. 1 discloses the process flow diagram of the vocal print verification system of preferred embodiment e-commerce transaction of the present invention.

Please refer to shown in Figure 1ly, the vocal print verification system of preferred embodiment e-commerce transaction of the present invention is at first logined customer account number when beginning to conclude the business by coupling arrangement.This coupling arrangement comprises personal computer (personal computer), an Automated Teller Machine (automated Teller Machine), contractor's machine for punching the card (credit card Verifier) etc., can connect to carry out the general commerce transaction.

Referring again to shown in Figure 1, then, client's basic document is sent to the vocal print authentication center, this vocal print authentication center can select to be arranged on contractor, financial institution or cura specialis mechanism.But vocal print authentication center utilization device for identifying is confirmed client's basic document, but but should comprise program identification logical circuit etc. by device for identifying.In addition, this vocal print authentication center has a vocal print verification system.

Referring again to shown in Figure 1, then, whether applied for vocal print relatively but should device for identifying check this client, whether promptly produce this client needs to carry out vocal print result relatively.This vocal print authentication center is passed this result back this coupling arrangement, so that carry out follow-up e-commerce transaction program.

Fig. 2 discloses the process block diagram of the vocal print verification system of preferred embodiment of the present invention.

Please refer to shown in Figure 2ly, the vocal print verification system 1 of preferred embodiment of the present invention comprises training system 10 and test macro 20, so that original input voice data is trained or test jobs.This vocal print verification system 1 comprises a front-end processing portion, a feature acquisition portion, a reservoir and an operational part in addition.This front-end processing portion and feature acquisition portion carry out front-end processing and feature acquisition for this training system 10 and test macro 20, and this reservoir is stored for phonetic feature, and this operational part is then with the in addition computing of this store voice feature and input phonetic feature.

When customer account number is imported vocal print verification system 1 of the present invention, can confirm identity.Then, this system is according to input account number Query Database, and whether this input account number belongs to is set up.When if this input account number is not set up, require whether enter this training system 10 and carry out the voice training operation, so that set up and store the voice data of this input account number.If when this input account number has been set up, enter this test macro 20 and carry out the tone testing operation, so that whether the phonetic feature of this input account number of identification meets the voice data that has stored this input account number.

Allow refer again to Figure 1 and Figure 2, then, when the client does not apply for that vocal print relatively, then enter and require the client to import personal identification number.If after the client imports incorrect personal identification number, promptly enter the refusal transactional stage.After importing correct personal identification number with the client, require whether to apply for the sound-groove identification registration.When selecting not apply for the sound-groove identification registration, promptly enter the permission transactional stage.Otherwise, when selecting the registration of application sound-groove identification, promptly enter the training system 10 of this vocal print verification system 1.

Details are as follows for this training system 10 of sound-groove identification of the present invention registration operation: before the acquisition phonetic feature, utilize this front-end processing portion with efficient voice information from original input voice DAQ, with the invalid voice messaging of filtering.The present invention's detecting comprises in short-term apart from energy (Short-Energy) and zero-crossing rate (zero-Crossing Rate).The present invention adopts the computing method in conjunction with Gauss's probability distribution, and its equation is as follows:

$b_i\stackrel{\→}{\left(X\right)}=\frac{1}{{\left(2\mathrm{\π}\right)}^{D/2}{\left|{\∑}_i\right|}^{1/2}}\exp \{-\frac{1}{2}{(\stackrel{\→}{x}-{\stackrel{\→}{u}}_i)}^1{\∑}_i^{-1}((\stackrel{\→}{x}-{\stackrel{\→}{u}}_i)\}---\left(1\right)$

Wherein

For original signal with its be divided into a plurality of D dimension the sound frame,

I=1 ..., M is affiliated probability, For expectation value, the ∑ i of ground unrest is the variance of ground unrest.At this, because

In D=256 be a definite value, so its omission will not be calculated, equation is simplified as follows:

$b_i\left(\stackrel{\→}{x}\right)=\frac{1}{{\left|{\∑}_i\right|}^{1/2}}\exp \{-\frac{1}{2}{(\stackrel{\→}{x}-{\stackrel{\→}{u}}_i)}^1{\∑}_i^{-1}(\stackrel{\→}{x}-{\stackrel{\→}{u}}_i)\}---\left(2\right)$

Exponent arithmetic in the following formula might be excessive on operational data, so after it is taken the logarithm, equation (2) simplification is as follows:

$b_i\left(\stackrel{\&RightArrow;}{x}\right)=\ln \left(\frac{1}{{\left|{\&Sum;}_i\right|}^{1/2}}\exp \right\{-\frac{1}{2}{(\stackrel{\&RightArrow;}{x}-{\stackrel{\&RightArrow;}{u}}_i)}^1{\&Sum;}_i^{-1}(\stackrel{\&RightArrow;}{x}-{\stackrel{\&RightArrow;}{u}}_i)\}<mtext>\; </mtext>\; )$

$=\mathrm{<figure-callout\; id="1"\; label="ln"\; filenames="A20051006318400131.png,A20051006318400141.png"\; state="\{\{state\}\}">ln</figure-callout>}\frac{1}{{\left|{\&Sum;}_i\right|}^{1/2}}-\frac{1}{2}{(\stackrel{\&RightArrow;}{x}-{\stackrel{\&RightArrow;}{u}}_i)}^1{\&Sum;}_i^{-1}(\stackrel{\&RightArrow;}{x}-{\stackrel{\&RightArrow;}{u}}_i)$

$b_i\left(\stackrel{\&RightArrow;}{x}\right)=(-\frac{1}{2})\ln \left|{\&Sum;}_i\right|--\frac{1}{2}{(\stackrel{\&RightArrow;}{x}-{\stackrel{\&RightArrow;}{u}}_i)}^1{\&Sum;}_i^{-1}(\stackrel{\&RightArrow;}{x}-{\stackrel{\&RightArrow;}{u}}_i)---\left(3\right)$

Capture former input voice data front end 256 points, calculate, then this two number and former input voice data this equation of substitution (3) are carried out computing in short-term apart from the expectation value and the variance of energy and zero-crossing rate.Utilize in short-term distribution probability differentiation efficient voice information and invalid voice messaging,, not only reduce data quantity, also can correctly capture efficient voice information the in addition filtering of invalid voice messaging apart from energy and zero-crossing rate.

Capture on the feature in this feature acquisition portion, the present invention adopts two special characteristic parameters of speech recognition, it comprises linear prediction cepstral coefficients (Linear Prediction Cepstrum Coefficient, LPCC) and Mel frequency marking cepstrum parameter (Mel Frequency Censtrum Coefficient, MFCC) both each 12 cepstrum parameters (cepstral coefficients) and 12 cepstrum parameters (delta-censtral coefficients).With the cepstrum parameter c _nTime is done partial differential

${\mathrm{\&Delta;c}}_n\left(t\right)=\frac{{\&PartialD;c}_n\left(t\right)}{\&PartialD;\left(t\right)}=\frac{\underset{k=-k}{\overset{k}{\&Sum;}}{\mathrm{kc}}_n(t+k)}{\underset{k=-k}{\overset{k}{\&Sum;}}{k}^2}---\left(4\right)$

K is for considering sound frame number.

Because the formula (4) of single order cepstrum parameter is too complicated, so it is simplified, following various when only considering each two time-frame of front and back, the equation simplification is as follows:

${\mathrm{\&Delta;c}}_n^0=[2*c(2,n)+c(1,n)]/5---\left(5\right)$

${\mathrm{\&Delta;<figure-callout\; id="1"\; label="c\; n"\; filenames="A20051006318400131.png,A20051006318400141.png"\; state="\{\{state\}\}">c</figure-callout>}}_n^{}=[2*c(3,n)+c(2,n)-c(0,n)]/6---\left(6\right)$

${\mathrm{\&Delta;c}}_n^i=[2*c(i+2,n)+c(i+1,n)-c(i-1,n)-2*c(i-2,n)]/10---\left(7\right)$

${\mathrm{\&Delta;c}}_n^{L-2}=[c(L-1,n)-c(L-3,n)-2*c(L-4,n)]/6---\left(8\right)$

${\mathrm{\&Delta;c}}_n^{L-1}=[-c(L-2,n)-2*c(L-3,n)]/5---\left(9\right)$

Equation (5) is in (9), and Cn is n rank eigenwerts, and L is a time-frame sum in the signal, and i is the time-frame numbering.

Fig. 3 discloses the synoptic diagram that concerns of the state of vocal print verification system of preferred embodiment of the present invention and sound frame.

On training utterance, voice have the idea of what is called " state ", the variation of mouth type and sound channel when state is pronunciation.Generally speaking, the mouth type of at every turn speaking necessarily changes, so the feature performance that each state all is voice to be changed.Sometimes a single-tone but might contain a plurality of states.A state also has fixed measure unlike the sound frame, a common state comprises a plurality of or dozens of sound frame.

Please refer to shown in Figure 3ly, first state comprises three sound frames, second state comprises six sound frames and the third state comprises four sound frames.

Fig. 4 discloses the sound frame of vocal print verification system of preferred embodiment of the present invention and the original allocation pattern diagram of state.This original allocation pattern three sample voice is for example divided equally action.

At originate mode voice are done to divide equally action, possibly can't divide exactly after dividing equally, unnecessary sound frame is then divided it equally at first and last state.Referring again to shown in Figure 3, in allocation model, sample voice is divided equally must consider 3 points: 1, the sound frame necessarily belongs to first state; 2, last sound frame necessarily belongs to last state; 3, the state variation of sound frame has only constant or is changed to the next one.Adopt Gauss to distribute probability to calculate each sound frame and belong to the probability of each state, and utilize viterbi algorithm to obtain the most similar path.

Fig. 5 discloses the state exchange synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

Please refer to shown in Figure 5, when three states, the direction that L sound frame may state exchange.Be considered as the state that can not belong to playing the fork tone frame, the direction of arrow is considered as may the state variation path.

Fig. 6 discloses the most similar path synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

Please refer to shown in Figure 6ly, the most similar path of acquisition feature has that first state comprises the 1st to 3 sound frame, second state comprises the 4th to 5 sound frame and the third state comprises the 7th to 10 sound frame.

Fig. 7 discloses the equal partial frame synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

With reference to shown in Figure 7, three sample voice are in the originate mode of three states, the distribution after it is adopted and divides equally.After each sample voice of first sample voice is divided equally three sound frames, remain two sound frames and divide configuration first state and second volt of attitude respectively.Each sample voice of second sample voice is divided equally four sound frames.After each sample voice of the 3rd sample voice is divided equally three sound frames, remain a sound frame and dispose first state respectively.After calculating, its maximal phase is 2157 like probability.

Redistribute sound frame synoptic diagram the first time that Fig. 8 discloses the vocal print verification system of preferred embodiment of the present invention.

Please refer to shown in Figure 8ly, after redistributing the sound frame for the first time, its maximal phase rises to 3171 like probability.

Redistribute sound frame synoptic diagram the second time that Fig. 9 discloses the vocal print verification system of preferred embodiment of the present invention.

Please refer to shown in Figure 9ly, after redistributing the sound frame for the second time, its maximal phase rises to 3571 like probability.

Figure 10 discloses the optimal allocation sound frame synoptic diagram of the vocal print verification system of preferred embodiment of the present invention.

Please refer to shown in Figure 10ly, after repeatedly redistributing the sound frame, its maximal phase no longer rises like probability 3571, so it is considered as optimal allocation sound frame.Calculate the expectation value of each state and variance as model parameter, this model parameter can be for being stored in database.

Referring again to shown in Figure 2, when entering this training system 10 and carry out the voice training operation, machine equation (1) to (9) obtains effective training utterance feature.Then utilize viterbi algorithm to obtain the most similar path.Then calculate the expectation value of each state and variance as model parameter, promptly finish the voice training operation.On voice training, its maximal phase, can't be tested voice training and finish the training operation during less than predetermined reference value like probability, thereby must operate this vocal print notarization system 1 again; Otherwise its maximal phase is like probability during greater than this predetermined reference value, by voice training, thereby model parameter is stored in this vocal print verification system 1.

Referring again to shown in Figure 1, when finishing the registration of application sound-groove identification, promptly enter the permission transactional stage.

Referring again to Figure 1 and Figure 2, when this input account number has been set up, enter this test macro 20 and carry out the tone testing operation.Details are as follows for this test macro 20 of sound-groove identification test operation of the present invention:

Same when entering this test macro 20 and carry out the tone testing operation, machine equation (1) to (9) obtains the Validity Test phonetic feature.

Referring again to shown in Figure 2, then, carry out between this tested speech feature of computing and the model parameter similar probability so that the output identification result.In speech recognition, its minimum similar probability by speech recognition, thereby can be left this vocal print verification system 1, and enter follow-up e-commerce transaction program during greater than predetermined reference value; Otherwise its minimum similar probability can't and finish test jobs by speech recognition, thereby must leave this vocal print verification system 1, and refusal is carried out follow-up e-commerce transaction program during less than this predetermined reference value.

Referring again to Figure 1 and Figure 2, at last, but should comply with the test macro 20 test results decision permission of this vocal print verification system 1 or refuse to carry out e-commerce transaction by device for identifying.

Claims (7)

1. An e-commerce transaction method, characterized in that it comprises steps: The customer account is logged in by a connected device; Use an identifiable device to confirm the basic information of the customer; The identifiable device checks whether a voiceprint comparison has been applied for; Using the voiceprint verification system for voiceprint identification; and The identifiable device decides to allow or deny the e-commerce transaction. 2. The e-commerce transaction method according to claim 1, wherein the voiceprint verification system includes: A front-end processing unit, which is used to perform front-end processing of the original input voice data of the voiceprint verification system, thereby completing the distinction between valid voice information and invalid voice information, and then extracting valid voice information; A feature extraction unit, which is used to extract the voice features of the effective voice information; a storage unit, which stores the voice feature; and A calculation unit, which calculates the stored voice features and the input voice features. 3. The e-commerce transaction method according to claim 2, wherein the voiceprint verification system further comprises a training system using the front-end processing unit and the feature extraction unit to obtain model parameters of the original input voice data. 4. The e-commerce transaction method according to claim 3, characterized in that the training system of the voiceprint verification system additionally uses the Viterbi algorithm to obtain the most similar path so as to calculate model parameters for storage. 5. The e-commerce transaction method according to claim 1, wherein the voiceprint verification system further comprises a test system using the front-end processing unit and the feature extraction unit to obtain voice features of the original input voice data. 6. The e-commerce transaction method as claimed in claim 1, characterized in that, when the voiceprint comparison is not applied, the voiceprint verification system enters the personal password. 7. The e-commerce transaction method according to claim 6, characterized in that, when the correct personal password is entered, it enters the stage of voiceprint identification whether to apply for registration.

Images