CN102306493A - Terminating machine, voice identification system and voice identification method thereof - Google Patents
Terminating machine, voice identification system and voice identification method thereof Download PDFInfo
- Publication number
- CN102306493A CN102306493A CN201110237569A CN201110237569A CN102306493A CN 102306493 A CN102306493 A CN 102306493A CN 201110237569 A CN201110237569 A CN 201110237569A CN 201110237569 A CN201110237569 A CN 201110237569A CN 102306493 A CN102306493 A CN 102306493A
- Authority
- CN
- China
- Prior art keywords
- parameter value
- voice
- value
- module
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Abstract
The invention relates to a terminating machine, a voice identification system and a voice identification method thereof. After the terminating machine successfully verifies a user voice, the information server receives and processes the user voice which is successfully verified and acquires a source target parameter value of the user voice successfully verified, and the source target parameter value is taken as a reference for the terminating machine to verify the user voice the next time, thus the terminating machine judges whether the current user is a valid user by taking the voice which is successfully verified the last time as the reference when the voice is verified while no voice content is required to be compared, and no password is required to be memorized; and identification degrees of two adjacent verified voices of a user are nearest, thus the situation that identification fault is caused by the fact that the voice is produced when the user has a cold or in a cracked voice can be avoided.
Description
Technical field
The present invention relates to the speech recognition technology, more specifically, relate to a kind of terminating machine, a kind of voice identification system and speech identifying method thereof.
Background technology
At present; People are that the password that input is set oneself is saved private data from damage; But password maybe be too tediously long or be made up of multiple coding (capitalization is English, small letter is English, numeral), As time goes on; The user might forget the password of setting originally, so usually can make the user be subjected to cause because of forgetting Password the puzzlement of password input error.And in state such as user flu or hoarseness following time, existing voice identification system is difficult to identify, and has brought very big inconvenience to the user.
Summary of the invention
A kind of voice identification system; This system comprises an information server and at least one terminating machine; The common checking of accomplishing input voice of this information server and at least one terminating machine; When this voice identification system is verified user speech success, carry out the parametrization processing and obtain this one first parameter value of verifying successful voice, one the 3rd parameter value and a source target component value; This voice identification system comprises a voice acquisition module, a characteristic acquisition module, a statistics probability module, a target acquisition module and a comparison module; This voice acquisition module is used to respond user's operation and obtains the voice of the current input of user; This characteristic acquisition module is used for obtaining Mei Er cepstrum feature MFCC from the voice of user's input; This statistical probability module is used for this MFCC that obtains and this first parameter value are carried out computing through the Podbielniak algorithm, obtains one the 4th parameter value; This target acquisition module is used for the 4th parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, obtains a voice target component value; And this comparison module is used for this voice target component value and this source target component value are compared; When the similarity of this voice target component value and this source target component value during more than or equal to a preset value; Confirm that this user's current speech verifies successfully; And, confirm this user's current speech authentication failed when the similarity of this voice target component value and this source target component value during less than this preset value.
A kind of terminating machine; This terminating machine and an information server communicate; This terminating machine comprises a voice-input unit, a processing unit and a storage unit; This voice-input unit is used to receive user's input voice; When this terminating machine was verified user speech success, this information server carried out the parametrization processing to these voice and obtains this one first parameter value of verifying successful voice, one the 3rd parameter value and a source target component value, and this this this first parameter value, the 3rd parameter value and this source target component value of verifying successful voice of terminating machine storage is in storage unit; This processing unit comprises a voice acquisition module, a characteristic acquisition module, a statistics probability module, a target acquisition module and a comparison module; This voice acquisition module is used for obtaining from this voice-input unit the voice of the current input of user; This characteristic acquisition module is used for obtaining from this user speech that obtains the Mei Er cepstrum feature MFCC of voice; This statistical probability module is used for first parameter value of this MFCC that obtains and this storage unit is carried out computing through the Podbielniak algorithm, obtains one the 4th parameter value; This target acquisition module is used for the 4th parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, and obtains a voice target component value; And the source target component value that this comparison module voice target component value of being used for this target acquisition module is obtained and this storage unit are stored compares; And the similarity of the source target component value of in voice target component value that this target acquisition module obtains and this storage unit, storing is during more than or equal to a preset value, confirms that this user's current speech verifies successfully.
A kind of speech identifying method; This method is applied to a voice identification system; The method comprising the steps of: when this voice identification system is verified user speech success, carry out the parametrization processing and obtain this one first parameter value of verifying successful voice, one the 3rd parameter value and a source target component value; The voice of the current input of user are obtained in response user's operation; From the user speech that this obtains, obtain Mei Er cepstrum feature MFCC; This MFCC that obtains and first parameter value are carried out computing through the Podbielniak algorithm, obtain one the 4th parameter value; The 4th parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, obtain a voice target component value; And the similarity of this voice target component value that obtains and this source target component value relatively; When this similarity during more than or equal to a preset value; Confirm that this user's current speech verifies successfully,, confirm this user's current speech authentication failed when this similarity during less than a predetermined value.
Voice identification system that the present invention relates to and method; After terminating machine is verified user speech success; This information server receives the source target component value of handling the successful user speech of this checking and obtaining the successful user speech of this checking; This source target component value is verified the reference of user speech next time as terminating machine; Thereby terminating machine is that the successful voice of above one-time authentication judge as a reference whether the active user is validated user when the checking voice, and need not compare voice content, does not need memory cipher; The identification of the checking voice that are close to for twice because of a user is immediate, so also avoided because the user is in the situation generation of the speech recognition error that states such as flu or hoarseness send.
Description of drawings
Fig. 1 is the structural representation of an embodiment of the present invention voice identification system.
Fig. 2 is the structural representation of the control module of Fig. 1 of the present invention.
Fig. 3 is the structural representation of the processing unit of Fig. 1 of the present invention.
Fig. 4 is the information server of Fig. 1 of the present invention and the hardware configuration synoptic diagram of terminating machine.
Fig. 5 and Fig. 6 are the method flow diagrams of the voice identification system speech recognition of Fig. 1 of the present invention.
The main element symbol description
| |
1 |
| |
10 |
| Terminating |
20 |
| |
11 |
| The |
111 |
| The |
112 |
| |
113 |
| The |
114 |
| The |
115 |
| The |
116 |
| |
117 |
| Voice- |
23 |
| |
24 |
| |
25 |
| |
241 |
| The |
242 |
| |
243 |
| |
244 |
| The |
245 |
| The |
246 |
| The |
247 |
| |
248 |
Following embodiment will combine above-mentioned accompanying drawing to further specify the present invention.
Embodiment
Please refer to Fig. 1-4, be the structural representation of the voice identification system in an embodiment of the present invention 1.This voice identification system 1 comprises an information server 10 and at least one terminating machine 20.This information server 10 communicates through wired or wireless mode with this at least one terminating machine 20.This information server 10 and the 20 common checkings of accomplishing input voice of at least one terminating machine.This at least one terminating machine 20 is used for comparing the checking of accomplishing current speech to the user speech of current reception and the successful voice of checking last time, after the current speech of this reception is verified successfully, sends these voice to information server 10.This information server 10 is used for the voice of receiving terminal machine 20, user speech is carried out parametrization handle, and obtains this user speech corresponding parameters value, and transmits this voice corresponding parameters value and accomplish the checking of user speech next time to terminating machine 20 references.
In one embodiment of the present invention, this information server 10 comprises a control module 11.This control module 11 comprises a voice receiver module 111, a characteristic acquisition module 112, a model building module 113, a statistics probability module 114, a matrix parameter module 115, a target acquisition module 116 and a delivery module 117.
This voice receiver module 111 is used to receive the user speech that transmits from least one terminating machine 20.This characteristic acquisition module 112 is used for obtaining from the voice of this reception the MFCC (Mel Frequency Cepstrum Coefficient, Mei Er cepstrum feature) of these voice.
This model building module 113 is used to obtain the MFCC that this characteristic acquisition module 112 obtains; Stray parameter value preset in the MFCC that this characteristic acquisition module 112 is obtained and this model building module is carried out computing through greatest hope algorithm (Expectation-maximization algorithm); Obtain the parameter value after the computing; With the maximization of the parameter value after this computing, and judge whether this maximization parameter value is a convergency value.When this maximization parameter value is not a convergency value; This model building module 113 continues to maximize parameter value and this MFCC eigenwert of obtaining is carried out computing after the current maximization parameter value that draws is a convergency value through the greatest hope algorithm, draws this maximization parameter value as first parameter value.Wherein, the stray parameter of this hypothesis can be initial weight value (Initial Weights), initial average output value (Initial Means) and initial variation value (Initial Variances).This output parameter value can be weighted value (Weights), mean value (Means) and variation value (Variances).Convergency value is meant that a back parameter value that obtains deducts behind the last parameter value ratio with back one parameter value less than a certain particular value, and wherein, this particular value is stipulated by the user.Parameter maximization is that this parameter value is parameter and maximizes after the back last relatively parameter value of one parameter value is restrained.
This statistical probability module 114 is used to obtain MFCC that this characteristic acquisition module 122 obtains and first parameter value of this model building module 113 outputs; And first parameter value of this MFCC eigenwert and this model building module 113 outputs carried out computing through Podbielniak algorithm (Baum-Welch algorithm), obtain one second parameter value.This second parameter value be single order and second order Baum-Welch algorithm statistical value (First And Second Order Baum Welch Statistics Of All Utterances Set) and maximum posterior probability (Posteriori Probability) wherein.
This matrix parameter module 115 is used to obtain first parameter value of these model building module 113 outputs and second parameter value of this statistical probability module 114 outputs; And second parameter value that first parameter value and this statistical probability module 114 of these model building module 113 outputs are exported carries out computing through total variation matrix algorithms (Total Variability Matrix); Draw a parameter value; And, judge whether this maximized parameter value is a convergency value with this parameter value maximization.When this maximization parameter value is not a convergency value; This matrix parameter module 115 continues to maximize parameter value and carries out computing with second parameter value that from this statistical probability module 114, obtains after this maximization parameter value is convergency value through total variation matrix algorithms, draws a convergent maximization parameter value as the 3rd parameter value.Wherein, the 3rd parameter value can be matrix parameter (T), mean value (Means) and the different matrix value of remaining co-variation (residual covariance Matrices).
This target acquisition module 116 is used to obtain the 3rd parameter value that second parameter value that this statistical probability module 114 draws and this matrix parameter module 115 draw; And this second parameter value and the 3rd parameter value carried out computing through characteristic parameter extraction algorithm (Factors Extraction), draw a source target component value.
This delivery module 117 is used to obtain the source target component value that first parameter value that this model building module 113 draws, the 3rd parameter value that this matrix parameter module 115 draws and this target acquisition module 116 draw, and those parameter values that this user speech is corresponding are sent to each terminating machine 20.
Therefore; In case information server 10 receives the user speech from terminating machine 20; Information server 10 is just handled this user speech and is drawn this user speech corresponding first parameter value, the 3rd parameter value and source target component value, and transmits each corresponding parameter value of this user speech to terminating machine 20.
This each terminating machine 20 comprises a voice-input unit 23, a processing unit 24 and a storage unit 25.This voice-input unit 23 is used to receive the user's voice input.In this embodiment, this voice-input unit 23 is a microphone.
This processing unit 24 comprises a reminding module 241, a voice acquisition module 242, a delivery module 243, a receiver module 244, a characteristic acquisition module 245, a statistics probability module 246, a target acquisition module 247 and a comparison module 248.This reminding module 241 is used to respond user's operation indicating user input voice.This voice acquisition module 242 is used to obtain the voice of user through voice-input unit 23 inputs.
This receiver module 244 is used to receive corresponding this first parameter value, the 3rd parameter value and this source target component value of each voice that this information server 10 transmits, and is stored in this storage unit 25 this first parameter value, the 3rd parameter value and this source target component value of said reception and last each parameter value that receives storage of deletion.
Therefore; When terminating machine 20 receives corresponding each parameter value of voice that information server 10 transmits; The checking that 20 pairs of these voice of this terminating machine are described is successful, stores each corresponding parameter value of these voice and in storage unit 25, reaches last each parameter value that receives storage of deletion.For example; During the n time checking of this terminating machine 20, one user speech; Suppose that this terminating machine 20 is successful at this user speech of (n-1) inferior checking, (n-1) inferior voice corresponding parameters value of this user that then this information server 10 of storage transmits in this storage unit 25.In this embodiment, can store the parameter value of multi-person speech in this storage unit 25.
This characteristic acquisition module 245 is used for obtaining MFCC from the user speech that voice acquisition module 242 obtains, and obtains the MFCC of voice.This statistical probability module 246 is used to obtain the MFCC that this characteristic acquisition module 245 obtains; And from this storage unit 25, obtain first parameter value; This MFCC that obtains and first parameter value are carried out computing through Podbielniak algorithm (Baum-Welch algorithm); Obtain one the 4th parameter value, like Jie and two Jie's Baum-Welch algorithm statistical value (First And Second Order Baum Welch Statistics Of All Utterances Set) and maximum posterior probability (Posteriori Probability).
This target acquisition module 247 is used to obtain the 4th parameter value that this statistical probability module 246 obtains, and from this storage unit 25, obtains the 3rd parameter value, and this 4th parameter value that obtains and the 3rd parameter value are carried out computing, obtains a voice target component value.This comparison module 248 is used for according to formula
Calculate the similarity of the source target component value of storing in voice target component value and this storage unit 25 of voice of user input.W wherein
TargetBe the source target component value of storage in this storage unit 25, w
TestThe voice target component value that obtains for these at least one terminating machine 20 computings.When similarity during more than or equal to a preset value, this comparison module 248 confirms that these user speech verify successfully.When similarity during less than this preset value, this comparison module 248 is confirmed these user speech authentication faileds.
When comparison module 248 confirms that this user speech is verified successfully; This delivery module 243 is used for sending the user speech that voice acquisition module 242 obtains to information server 10 to be handled; This information server 10 is carried out aforesaid operation, obtains first parameter value, the 3rd parameter value and the source target component value of the successful user speech of this checking.This information server 10 is sent to terminating machine 20 storages to each parameter value of this user speech, and therefore, each parameter value of storage is about the successful voice of the last checking of user in the storage unit 25 of terminating machine 20.When this terminating machine 20 next time during speech verification, judge the legitimacy of voice next time through comparing with the successful voice of the last time checking.
When the user uses these terminating machine 20 checkings first; The user that the delivery module 243 of terminating machine 20 obtains voice acquisition module 242 voice first sends information server 10 to and handles; This information server 10 is carried out aforesaid operation, and obtaining first, first parameter value, the 3rd parameter value and the source target component value of voice arrive terminating machine 20 storages.
In one second embodiment of the present invention, the user speech of 10 pairs of receptions of information server carries out parametrization to be handled and obtains this user speech corresponding parameters value and can be accomplished by terminating machine 20.In one the 3rd embodiment of the present invention; This at least one terminating machine 20 is the user speech of current reception and verified that successful voice compared the checking of accomplishing current speech and can be accomplished by information server 10 last time, and this information server 10 sends to relevant terminal machine 20 to the checking result.
Therefore; After the success of terminating machine 20 checkings one user speech; This information server 10 receives the source target component value of handling the successful user speech of this checking and obtaining the successful user speech of this checking; This source target component value is verified the reference of user speech next time as terminating machine 20; Thereby terminating machine 20 is that the successful voice of above one-time authentication judge as a reference whether the active user is validated user when the checking voice, and need not compare voice content, does not need memory cipher; The identification of the checking voice that are close to for twice because of a user is immediate, so also avoided because the user is in the situation generation of the speech recognition error that states such as flu or hoarseness send.
Please refer to Fig. 5 and Fig. 6, be the method flow diagram of the voice identification system speech recognition of Fig. 1.
In step S501, this reminding module 241 response users' operation indicating user input voice.In step S502, this voice acquisition module 242 obtains the voice of user's input.In step S503, this characteristic acquisition module 245 obtains MFCC from the voice of user's input.
In step S504; This statistical probability module 246 is obtained the MFCC that this characteristic acquisition module 245 obtains; And from this storage unit 25, obtain this first parameter value; This MFCC eigenwert of obtaining and this first parameter value are carried out computing through Podbielniak algorithm (Baum-Welch algorithm), and obtain one the 4th parameter value.
In step S505; This target acquisition module 247 obtains the 4th parameter value that statistical probability module 237 obtains; And from this storage unit 25, obtain the 3rd parameter value; And the 4th parameter value that this statistical probability module 246 is obtained calculates through characteristic parameter extraction algorithm (Factors Extraction) with the 3rd parameter value, obtains a voice target component.
In step S506, the comparison module 248 of this terminating machine 20 is used for according to formula
The voice target component value and this that calculate user input voice are stored in the similarity of the source target component value in the storage unit 25, and when similarity during more than or equal to this preset value, confirm this speech verification success.Wherein, w
TargetBe the source target component value of storing in this storage unit, w
TestThe voice target component value that obtains at least one terminating machine computing.When similarity during less than this preset value, this speech verification failure, this flow process finishes.
When this speech verification success, in step S601, the voice receiver module 111 receiving terminal machines 20 of this information server 10 are through the user speech of voice-input unit 23 inputs.In step S602, this characteristic acquisition module 112 obtains MFCC from the user speech that this receives.
In step S603; This model building module 113 obtains the MFCC that this characteristic acquisition module 112 obtains; And stray parameter value preset in this MFCC and this model building module 113 carried out computing through greatest hope algorithm (Expectation-maximization algorithm); Thereby obtain a parameter value,, and judge whether this maximized parameter value is a convergency value this parameter value maximization.When this parameter value was not a convergency value, it was a convergency value until this parameter value that the MFCC eigenwert that this parameter value repeats and this characteristic acquisition module obtains is done calculating, obtains one first parameter value.
In step S604; This statistical probability module 114 is obtained first parameter value that this MFCC eigenwert and this model building module 113 obtain; And first parameter value that this MFCC eigenwert and this model building module 113 obtain carried out computing through Podbielniak algorithm (Baum-Welch algorithm); Thereby obtain one second parameter value; Wherein, this second parameter value can be Baum-Welch algorithm statistical value (First And Second Order Baum Welch Statistics Of All Utterances Set) and the maximum posterior probability (Posteriori Probability) that Jie and two is situated between.
In step S605; This matrix parameter module 115 is obtained second parameter value that first parameter value that this model building module 113 obtains and this statistical probability module 114 obtain; And second parameter value that first parameter value that this model building module 113 is obtained and this statistical probability module 114 obtain carries out computing through total variation matrix algorithms (Total Variability Matrix); Draw a parameter value,, and judge whether this maximization parameter value restrains the parameter value maximization.When this maximization parameter value is not restrained, this maximization parameter value is calculated with the parameter value that this statistical probability module of obtaining obtains again, until the parameter value convergence, obtain one the 3rd parameter value.
In step S606; This target acquisition module 116 obtains the 3rd parameter value that second parameter value that this statistical probability module 114 obtains and this matrix parameter module 115 obtain; And the 3rd parameter value that second parameter value that this statistical probability module 114 is obtained and this matrix parameter module 115 obtain carries out computing through characteristic parameter extraction algorithm (Factors Extraction), obtains a source target component value.
In step S607, the source target component value that first parameter value that this delivery module 117 is used for this model building module 113 is obtained, the 3rd parameter value that matrix parameter module 115 obtains and target acquisition module 116 obtain is sent to each terminating machine 20.
In step S608; Said each parameter value of these terminating machine 20 storages last each parameter value that receives storage in storage unit 25 and in the deletion storage unit 25; Thereby this terminating machine 20 is stored in each up-to-date in the storage unit 25 parameter value with reference to said when verifying user speech next time.
Those skilled in the art will be appreciated that; Above embodiment only is to be used for explaining the present invention; And be not to be used as qualification of the present invention; As long as within connotation scope of the present invention, appropriate change that above embodiment did is all dropped within the scope that the present invention requires to protect with changing.
Claims (7)
1. voice identification system, this system comprises an information server and at least one terminating machine, the common checking of accomplishing input voice of this information server and at least one terminating machine is characterized in that:
When this voice identification system is verified user speech success, carry out the parametrization processing and obtain this one first parameter value of verifying successful voice, one the 3rd parameter value and a source target component value;
This voice identification system comprises a voice acquisition module, a characteristic acquisition module, a statistics probability module, a target acquisition module and a comparison module;
This voice acquisition module is used to respond user's operation and obtains the voice of the current input of user;
This characteristic acquisition module is used for obtaining Mei Er cepstrum feature MFCC from the voice of user's input;
This statistical probability module is used for this MFCC that obtains and this first parameter value are carried out computing through the Podbielniak algorithm, obtains one the 4th parameter value;
This target acquisition module is used for the 4th parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, obtains a voice target component value; And
This comparison module is used for this voice target component value and this source target component value are compared; When the similarity of this voice target component value and this source target component value during more than or equal to a preset value; Confirm that this user's current speech verifies successfully; And, confirm this user's current speech authentication failed when the similarity of this voice target component value and this source target component value during less than this preset value.
2. voice identification system as claimed in claim 1; It is characterized in that: this information server comprises a control module, and this control module comprises a voice receiver module, a characteristic acquisition module, a model building module, a statistics probability module, a matrix parameter module, a target acquisition module and a delivery module;
The voice receiver module of this control module receives the successful user speech of checking that this at least one terminating machine transmits;
The characteristic acquisition module of this control module is used for obtaining from the user speech of this reception the Mei Er cepstrum feature MFCC of user speech;
The model building module of this control module is used for the MFCC of voice and a preset stray parameter value are carried out computing through the greatest hope algorithm; Draw a parameter value; With this parameter value maximization, and when this maximization parameter value is a convergency value, obtain this first parameter value;
The statistical probability module of this control module is used for this MFCC eigenwert and this first parameter value are carried out computing through the Podbielniak algorithm, and obtains one second parameter value;
The matrix parameter module of this control module is used for this first parameter value and this second parameter value are carried out computing through total variation matrix algorithms, draws a parameter value, with this parameter value maximization, and when this maximization parameter value convergence, obtains the 3rd parameter value;
The target acquisition module of this control module is used for this second parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, obtains a source target component value; And
The delivery module of this control module is used for first parameter value, and the 3rd parameter value and this source target component value are sent to this at least one terminating machine.
3. voice identification system as claimed in claim 2 is characterized in that: this model building module also is used for when this parameter value is not a convergency value, and this parameter value is continued to carry out computing with the MFCC of these voice.
4. voice identification system as claimed in claim 2 is characterized in that: this matrix parameter module also is used for when this parameter value is not a convergency value, and second parameter value that this parameter value is continued to obtain with the statistical probability module carries out computing.
5. terminating machine, this terminating machine and an information server communicate, and this terminating machine comprises a voice-input unit, a processing unit and a storage unit, and this voice-input unit is used to receive user's input voice, it is characterized in that:
When this terminating machine is verified user speech success; This information server carries out the parametrization processing to these voice and obtains this one first parameter value of verifying successful voice, one the 3rd parameter value and a source target component value, and this this this first parameter value, the 3rd parameter value and this source target component value of verifying successful voice of terminating machine storage is in storage unit;
This processing unit comprises a voice acquisition module, a characteristic acquisition module, a statistics probability module, a target acquisition module and a comparison module;
This voice acquisition module is used for obtaining from this voice-input unit the voice of the current input of user;
This characteristic acquisition module is used for obtaining from this user speech that obtains the Mei Er cepstrum feature MFCC of voice;
This statistical probability module is used for first parameter value of this MFCC that obtains and this storage unit is carried out computing through the Podbielniak algorithm, obtains one the 4th parameter value;
This target acquisition module is used for the 4th parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, and obtains a voice target component value; And
The source target component value that the voice target component value that this comparison module is used for this target acquisition module is obtained and this storage unit are stored compares; And the similarity of the source target component value of in voice target component value that this target acquisition module obtains and this storage unit, storing is during more than or equal to a preset value, confirms that this user's current speech verifies successfully.
6. speech identifying method, this method is applied to a voice identification system, it is characterized in that, and the method comprising the steps of:
When this voice identification system is verified user speech success, carry out the parametrization processing and obtain this one first parameter value of verifying successful voice, one the 3rd parameter value and a source target component value;
The voice of the current input of user are obtained in response user's operation;
From the user speech that this obtains, obtain Mei Er cepstrum feature MFCC;
This MFCC that obtains and first parameter value are carried out computing through the Podbielniak algorithm, obtain one the 4th parameter value;
The 4th parameter value and the 3rd parameter value are carried out computing through the characteristic parameter extraction algorithm, obtain a voice target component value; And
The similarity of this voice target component value that obtains and this source target component value relatively; When this similarity during more than or equal to a preset value; Confirm that this user's current speech verifies successfully,, confirm this user's current speech authentication failed when this similarity during less than a predetermined value.
7. speech identifying method as claimed in claim 6 is characterized in that, this method also comprises:
Obtain the successful user speech of checking;
From this user speech, obtain the MFCC of user speech;
The MFCC of these voice is carried out computing with preset stray parameter value through the greatest hope algorithm, draw a parameter value,, and when this parameter value is a convergency value, draw this first parameter value the parameter value maximization of computing;
This MFCC eigenwert and first parameter value are carried out computing through the Podbielniak algorithm, obtain one second parameter value;
This first parameter value and this second parameter value are carried out computing through total variation matrix algorithms, draw a parameter value,, and when this maximization parameter value convergence, export the 3rd parameter value the parameter value maximization; And
This second parameter value and the 3rd parameter value are calculated through the characteristic parameter extraction algorithm, draw a source target component value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110237569A CN102306493A (en) | 2011-08-18 | 2011-08-18 | Terminating machine, voice identification system and voice identification method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110237569A CN102306493A (en) | 2011-08-18 | 2011-08-18 | Terminating machine, voice identification system and voice identification method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102306493A true CN102306493A (en) | 2012-01-04 |
Family
ID=45380343
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110237569A Withdrawn CN102306493A (en) | 2011-08-18 | 2011-08-18 | Terminating machine, voice identification system and voice identification method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102306493A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103448632A (en) * | 2012-05-28 | 2013-12-18 | 百度在线网络技术（北京）有限公司 | Automobile control method and device |
| CN104036168A (en) * | 2014-06-06 | 2014-09-10 | 北京智谷睿拓技术服务有限公司 | Authentication method and equipment |
| CN109960910A (en) * | 2017-12-14 | 2019-07-02 | 广东欧珀移动通信有限公司 | Method of speech processing, device, storage medium and terminal device |
| CN112667978A (en) * | 2020-12-31 | 2021-04-16 | 深圳市创奇电气有限公司 | High-voltage switch cabinet remote monitoring method and system and storage medium |
-
2011
- 2011-08-18 CN CN201110237569A patent/CN102306493A/en not_active Withdrawn
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103448632A (en) * | 2012-05-28 | 2013-12-18 | 百度在线网络技术（北京）有限公司 | Automobile control method and device |
| CN103448632B (en) * | 2012-05-28 | 2015-11-25 | 百度在线网络技术（北京）有限公司 | Automobile control method and device |
| CN104036168A (en) * | 2014-06-06 | 2014-09-10 | 北京智谷睿拓技术服务有限公司 | Authentication method and equipment |
| CN109960910A (en) * | 2017-12-14 | 2019-07-02 | 广东欧珀移动通信有限公司 | Method of speech processing, device, storage medium and terminal device |
| CN109960910B (en) * | 2017-12-14 | 2021-06-08 | Oppo广东移动通信有限公司 | Voice processing method, device, storage medium and terminal equipment |
| CN112667978A (en) * | 2020-12-31 | 2021-04-16 | 深圳市创奇电气有限公司 | High-voltage switch cabinet remote monitoring method and system and storage medium |
| CN112667978B (en) * | 2020-12-31 | 2023-03-24 | 深圳市创奇电气有限公司 | High-voltage switch cabinet remote monitoring method and system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3195311B1 (en) | Method and apparatus for performing speaker recognition | |
| GB2529503B (en) | Voice authentication system and method | |
| US9607621B2 (en) | Customer identification through voice biometrics | |
| US9373330B2 (en) | Fast speaker recognition scoring using I-vector posteriors and probabilistic linear discriminant analysis | |
| US8234494B1 (en) | Speaker-verification digital signatures | |
| US10276168B2 (en) | Voiceprint verification method and device | |
| US11037574B2 (en) | Speaker recognition and speaker change detection | |
| CA2736133C (en) | Voice authentication system and methods | |
| KR20190015488A (en) | Voice user interface | |
| CN106961418A (en) | Identity identifying method and identity authorization system | |
| WO2014166362A1 (en) | Method, server, client and system for verifying verification codes | |
| CN102306493A (en) | Terminating machine, voice identification system and voice identification method thereof | |
| US10630680B2 (en) | System and method for optimizing matched voice biometric passphrases | |
| CN105205367B (en) | Information processing method and electronic equipment | |
| CN111816184B (en) | Speaker recognition method, speaker recognition device, and recording medium | |
| EP4055592A1 (en) | Systems and methods for customer authentication based on audio-of-interest | |
| US20140095169A1 (en) | Voice authentication system and methods | |
| US20120330663A1 (en) | Identity authentication system and method | |
| US20190130920A1 (en) | Method and apparatus for training voiceprint recognition system | |
| US9773150B1 (en) | Method and system for evaluating fingerprint templates | |
| CN105827578A (en) | Method and device for processing keystroke feature data | |
| CN114974212A (en) | Method for speech recognition and apparatus using the same | |
| CN110059474B (en) | PIN code modification method and device | |
| US9231924B2 (en) | Gesture-initiated encryption using error correction coding | |
| US20230153815A1 (en) | Methods and systems for training a machine learning model and authenticating a user with the model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C04 | Withdrawal of patent application after publication (patent law 2001) | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20120104 |
|
| C04 | Withdrawal of patent application after publication (patent law 2001) | ||
| WW01 | Invention patent application withdrawn after publication |