CN106356066A - Speech recognition system based on cloud computing - Google Patents

Abstract

The invention provides a speech recognition system based on cloud computing. The speech recognition system comprises a speech input module, a load balancing control module, a speech recognition platform and a speech cloud database, wherein the speech input module is used for inputting to-be-recognized speech data and transmitting the speech data to the load balancing control module; the speech recognition platform comprises multiple speech recognition servers, and the load balancing control module is used for dynamically allocating speech recognition tasks to idle speech recognition servers for recognition according to load of the speech recognition servers; the speech cloud database is used for storing a speech library template of a hidden Markov model, and the speech recognition servers quickly compare the to-be-recognized speech data with data in the speech cloud database and return recognition results to users. With the adoption of the speech recognition system, the problems of low recognition rate, slow access response and the like of speech recognition systems can be effectively solved, the comparison running time is prolonged, and the adaptability is wide.

Description

A kind of speech recognition system based on cloud computing

Technical field

The present invention relates to technical field of voice recognition is and in particular to a kind of speech recognition system based on cloud computing.

Background technology

In correlation technique, speech recognition technology can be only applied to limited scope, this is because, due to hardware device, number According to the restriction of the factors such as amount of storage, region, once data increase is increased by a lot of run times, remote speech identification is difficult to real Now the suitability of therefore speech recognition technology is very narrow.

Content of the invention

For solving the above problems, the present invention is intended to provide a kind of speech recognition system based on cloud computing.

The purpose of the present invention employs the following technical solutions to realize:

Provide a kind of speech recognition system based on cloud computing, including voice input module, load balancing control module, Speech recognition platforms, speech cloud data base, voice input module is connected with load balancing control module, load balancing control module It is connected with speech recognition platforms, speech recognition platforms are connected with speech cloud data base；Described voice input module is used for input to be needed Speech data to be identified, and speech data is sent to described load balancing control module；Described speech recognition platforms include Multiple speech recognition servers, described load balancing control module is used for speech recognition according to the load of speech recognition server Task dynamically distributes are identified to idle speech recognition server；Described speech cloud data base is used for storing hidden Ma Erke The voice library template of husband's model, the speech data needing identification is entered by speech recognition server with the data in speech cloud data base Row is quick to be compared, and is that user returns recognition result in time.

The invention has the benefit that can the discrimination that exists of the speech recognition system in effectively solving correlation technique low, The problems such as storage and the access response that causes of computing resource anxiety are slow, not only increases contrast run time, and the suitability is wide, Thus solving above-mentioned technical problem.

Brief description

Using accompanying drawing, the invention will be further described, but the embodiment in accompanying drawing does not constitute any limit to the present invention System, for those of ordinary skill in the art, on the premise of not paying creative work, can also obtain according to the following drawings Other accompanying drawings.

Fig. 1 is present configuration connection diagram.

Fig. 2 is the structural representation of data safety management system of the present invention.

Reference:

Voice input module 1, load balancing control module 2, speech recognition platforms 3, data safety management system 4, voice Cloud data base 5, data service system 40, data pretreatment 41, cloud storage encrypting and deciphering system 42, control system 43, safety Administrative center 44.

Specific embodiment

The invention will be further described with the following Examples.

Application scenarios 1

Referring to Fig. 1, Fig. 2, the speech recognition system based on cloud computing of an embodiment of this application scene, including voice Input module 1, load balancing control module 2, speech recognition platforms 3, speech cloud data base 5, voice input module 1 is with load all Weighing apparatus control module 2 connects, and load balancing control module 2 is connected with speech recognition platforms 3, speech recognition platforms 3 and speech cloud number Connect according to storehouse 5；Described voice input module 1 is used for input needs the speech data of identification, and speech data is sent to described Load balancing control module 2；Described speech recognition platforms 3 include multiple speech recognition servers, and described load balancing controls mould Block 2 is for the load according to speech recognition server by voice recognition tasks dynamically distributes to idle speech recognition server It is identified；Described speech cloud data base 5 is used for storing the voice library template of HMM, and speech recognition server will The speech data needing identification is quickly compared with the data in speech cloud data base 5, is that user returns identification knot in time Really.、

Preferably, described speech recognition platforms 3 include at least two voices knowledges that can separately provide speech identifying function Other server.

The above embodiment of the present invention can the discrimination that exists of the speech recognition system in effectively solving correlation technique be low, storage The problems such as slow with the access response that computing resource anxiety causes, not only increase contrast run time, and the suitability is wide, thus Solve above-mentioned technical problem.

Preferably, connected with the pattern that cloud net connects between speech recognition server.

Connected with the pattern that cloud net connects between this preferred embodiment speech recognition server, improve speech-recognition services The access response speed of device.

Preferably, the described speech recognition system based on cloud computing is also included for carrying out safety management to speech data Data safety management system 4；Described data safety management system 4 includes data service system 40, data pretreatment 41, cloud Storage encrypting and deciphering system 42, control system 43 and security management center 44；Described data service system 40 is used for being responsible for voice number According to storage management, backup and inquire about；The speech data that described data pretreatment 41 is used for need maintain secrecy carries out pretreatment； Described cloud storage encrypting and deciphering system 42 is used for the speech data that need maintain secrecy being carried out add according to the access control safety strategy optimizing Close or deciphering；Described control system 43 is used for storing speech data to corresponding storage device；Described security management center 44 For unified monitoring management is carried out safely to each system.

This preferred embodiment constructs the system structure of data safety management system 4.

Preferably, storage, backup and the inquiry of described responsible speech data, comprising:

(1) data form is changed, set up and be applied to the form that non-relational database is stored；

(2) basic data and expert data are splitted data into, the strategy using centralized and distributed combination enters to data Row storage, during storage, all data are all backed up；The strategy inclusion of described centralized and distributed combination: for higher than default The basic data of frequency adopts centralised storage, adopts distributed storage for the expert data less than predeterminated frequency；

(3) set up corresponding data retrievad algorithm, data is carried out with quick-searching, described data retrievad algorithm adopts catalogue The mode that retrieval and search engine combine is carried out, and specifically includes: set up data directory, according to catalogue, data is tentatively examined Rope；Input key word in search engine, precise search is carried out to data；Search engine finds the number of coupling according to certain mode According to, and be ranked up feeding back to user according to the matching degree of data and key word.

This preferred embodiment adopts the searching algorithm that catalogue retrieval and search engine combine, and can fast and accurately obtain Data.

Preferably, described unified monitoring management is carried out safely to each system, comprising:

(1) it is directed to data service system 40, data pretreatment 41, cloud storage encrypting and deciphering system 42, control system 43 Different security protections requires to take corresponding safety protection technique, is equipped with related safety protection equipment, forms complete peace Full protection system；

(2) set up effective Data Security, the safety in data storage, transmission, access process is carried out with comprehensive examining Consider, not only data is encrypted, the host-host protocol of data is encrypted simultaneously；

(3) set up virus and wooden horse defense mechanism, regularly update virus base and upgrading fire wall, the update cycle is t, and t takes It is worth for 6-10 days, the abnormal data detecting will be analyzed, and send early warning.

This preferred embodiment achieves the unified monitoring management to each system safety.

Preferably, described data pretreatment 41 includes data partitioning unit, data pick-up unit and access control peace Full policy optimization unit, described data partitioning unit is used for the speech data of need secrecy is divided into the data set of multiple mutual exclusions Close；Described data pick-up unit is used for the data acquisition system of described mutual exclusion is ranked up according to self-defining ordering rule, will be every First data cell in individual data acquisition system sequentially extracts, and preserves as small block data together with described ordering rule, There is not any association between the data cell two-by-two that wherein said mutual exclusion represents in data acquisition system；Described access control safety plan Slightly optimize the access control that unit is used for the access control safety policy optimization method generation system based on fine granularity division of resources Security strategy, comprising:

(1) data acquisition system based on the mutual exclusion after data pick-up cell processing, builds hierarchical data table structure, described Hierarchical data tree construction is three layer data tree constructions, and it includes service layer, logical layer and physical layer, and described service layer is and number According to the related root vertex of dispatch service, described logical layer is the data of association in access control safety strategy, described physical layer Comprise the data cell in the data acquisition system of all mutual exclusions；

(2) access control safety of the data for different safety class is formulated based on access control markup language xacml Strategy, the rule with data association in access control safety strategy is projected to the data cell in the data acquisition system of described mutual exclusion On, thus the rule in access control safety strategy is refine to data dimension；

(3) the enterprising line discipline of the data cell in the data acquisition system of each described mutual exclusion optimization, to delete distribution every The conflict of the rule in individual data cell and redundancy；

(4) merge the rule after optimizing, generate the access control safety strategy optimizing.

Preferably, described by data storage to corresponding storage device, comprising:

(1) small block data is stored to local storage, and using user-defined encryption technology, small block data is carried out Encryption；

(2) store in described cloud storage module 3 after remaining data being encrypted by cloud storage encrypting and deciphering system 42；Its In, after cloud storage module 3 receives data, cloud carries out to this data being saved in memory node after completeness check.

Above-mentioned two preferred embodiment arranges data pretreatment 41, first carries out data to the speech data of need secrecy and divides Cut data extraction to process, then the rule refinement controlling in security strategy that conducts interviews, it is possible to reduce the physics of data storage is deposited Storage space, reduces the expense of storage, and eliminates the conflict in access control safety strategy and redundancy, improves access control decision effect Rate；Extraction partial data is processed by data pick-up and stores in local storage, remainder data arranges corresponding access control Store after security strategy to cloud storage module 3, solve traditional cloud storage data-privacy based on simple encryption technology and protect The larger overhead of ratio that barrier mechanism is brought in actual process operation data and loaded down with trivial details, can effectively prevent malicious user Or cloud storage manager illegally steals, distorts the private data of user, improve the safety of the speech data storage that need to maintain secrecy Energy.

Preferably, described cloud storage encrypting and deciphering system 42 main by data owner, attribute mechanism, cloud, credible tripartite, use Five, family entity is constituted, and the described speech data to need secrecy is encrypted or deciphers, comprising:

(1) credible tripartite is user and attribute mechanism is respectively allocated User Identity uaid and attribute authority identity mark Aid, comprising:

A, initialized, credible tripartite's initialization system parameter isWherein α is random integers；

B, for each validated user, credible tripartite distribution uaid simultaneously Generates Certificate for it:

$certificate\left(uaid\right)=\hat{e}{\left(h\right(uaid),g)}^{c_{uaid}}$

Meanwhile, announce the authentication parameter of validated userWherein, c_uaid∈z_p；

C, generate identity key pair for data owner and validated user；

(2) generate the encryption and decryption key of identity-based, attribute encryption and decryption key and act on behalf of re-encrypted private key, wherein said The encryption and decryption key of identity-based includes identity public key gk_uaidWith identity private key ck_uaid, described attribute encryption and decryption key include belong to Property public key gk_aidWith attribute private key ck_aid:

${\mathrm{gk}}_{uaid}=\hat{e}{(g,g)}^{{\&proportional;}_{aid}}$

${\mathrm{gk}}_{aid}=\{\∀x\&element;{\mathrm{as}}_{aid}:{\mathrm{gk}}_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}}\}$

ck_uaid=(∝_aid,β_aid)

${\mathrm{ck}}_{aid}=(k_0=g^{{\&proportional;}_{aid}}{g}^{\mathrm{\α}\mathrm{\γ}},k_1=g^{\mathrm{\α}\mathrm{\γ}},\∀x\&element;{\mathrm{as}}_{uaid,aid}:k_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}\mathrm{\γ}})$

Wherein, as_aidThe community set that can distribute for single attribute mechanism, gk_xFor the public key of attribute x, b_xFor attribute x's Version number, ∝_aidFor the private key parameter of attribute mechanism, β_aidFor attribute undated parameter, as_uaid,aidIt is the identity according to attribute mechanism The community set of distribution, the parameter that γ randomly chooses for attribute mechanism, γ, ∝_aid,β_aid∈z_p；

(3) cloud storage encrypting and deciphering system 42 carries out data encryption using data key to speech data, obtains ciphertext ct, so After be utilized respectively identity public key and attribute public key to data key encrypt, generate identity key ciphertext ct_uWith attribute key ciphertext ct_a, comprising:

Character string ik of a, at random two regular lengths of generation, ak, merge and generate data key dk:

Dk=ik | | ak

B, carry out data encryption using the data that data key dk stores cloud storage module 3 to need, after obtaining ciphertext ct, Using attribute public key, ak is encrypted, generate attribute key ciphertext ct_a, using identity public key, ik is encrypted, generates identity key close Civilian ct_u；

(4) carry out acting on behalf of re-encryption, when receiving the request of data of user, cloud is close by identity using acting on behalf of re-encrypted private key Key ciphertext ct_uIt is converted into the ciphertext that specified user can decipher, wherein said re-encrypted private key of acting on behalf of uses itself by data owner Private key and identity public key calculate and generate；

(5), when carrying out data deciphering, after user receives data, it is utilized respectively identity private key ck_uaidWith attribute private key ck_aid Decryption identity key ciphertext ct_uWith attribute key ciphertext ct_a, then reconstruct data key, decrypting ciphertext ct；

(6) carry out the renewal of attribute and identity key.

This preferred embodiment passes through to arrange cloud storage encrypting and deciphering system 42, is capable of the fine granularity to eurypalynous data Access control and secret protection, resist user and the collusion of attribute mechanism simultaneously；Speech data to need secrecy, constructs respectively and is based on The encryption and decryption key of identity, attribute encryption and decryption key, merge composition data encryption key and this data are encrypted, thus only Meet identity simultaneously and the user of attribute double condition can decipher, greatly improve the safety of data safety management system 4 Energy.

In this application scenarios, update cycle t takes 6, and the safety of system improves 12% relatively.

Application scenarios 2

Referring to Fig. 1, Fig. 2, the speech recognition system based on cloud computing of an embodiment of this application scene, including voice Input module 1, load balancing control module 2, speech recognition platforms 3, speech cloud data base 5, voice input module 1 is with load all Weighing apparatus control module 2 connects, and load balancing control module 2 is connected with speech recognition platforms 3, speech recognition platforms 3 and speech cloud number Connect according to storehouse 5；Described voice input module 1 is used for input needs the speech data of identification, and speech data is sent to described Load balancing control module 2；Described speech recognition platforms 3 include multiple speech recognition servers, and described load balancing controls mould Block 2 is for the load according to speech recognition server by voice recognition tasks dynamically distributes to idle speech recognition server It is identified；Described speech cloud data base 5 is used for storing the voice library template of HMM, and speech recognition server will The speech data needing identification is quickly compared with the data in speech cloud data base 5, is that user returns identification knot in time Really.、

Preferably, described speech recognition platforms 3 include at least two voices knowledges that can separately provide speech identifying function Other server.

The above embodiment of the present invention can the discrimination that exists of the speech recognition system in effectively solving correlation technique be low, storage The problems such as slow with the access response that computing resource anxiety causes, not only increase contrast run time, and the suitability is wide, thus Solve above-mentioned technical problem.

Preferably, connected with the pattern that cloud net connects between speech recognition server.

Connected with the pattern that cloud net connects between this preferred embodiment speech recognition server, improve speech-recognition services The access response speed of device.

Preferably, the described speech recognition system based on cloud computing is also included for carrying out safety management to speech data Data safety management system 4；Described data safety management system 4 includes data service system 40, data pretreatment 41, cloud Storage encrypting and deciphering system 42, control system 43 and security management center 44；Described data service system 40 is used for being responsible for voice number According to storage management, backup and inquire about；The speech data that described data pretreatment 41 is used for need maintain secrecy carries out pretreatment； Described cloud storage encrypting and deciphering system 42 is used for the speech data that need maintain secrecy being carried out add according to the access control safety strategy optimizing Close or deciphering；Described control system 43 is used for storing speech data to corresponding storage device；Described security management center 44 For unified monitoring management is carried out safely to each system.

This preferred embodiment constructs the system structure of data safety management system 4.

Preferably, storage, backup and the inquiry of described responsible speech data, comprising:

(1) data form is changed, set up and be applied to the form that non-relational database is stored；

(2) basic data and expert data are splitted data into, the strategy using centralized and distributed combination enters to data Row storage, during storage, all data are all backed up；The strategy inclusion of described centralized and distributed combination: for higher than default The basic data of frequency adopts centralised storage, adopts distributed storage for the expert data less than predeterminated frequency；

(3) set up corresponding data retrievad algorithm, data is carried out with quick-searching, described data retrievad algorithm adopts catalogue The mode that retrieval and search engine combine is carried out, and specifically includes: set up data directory, according to catalogue, data is tentatively examined Rope；Input key word in search engine, precise search is carried out to data；Search engine finds the number of coupling according to certain mode According to, and be ranked up feeding back to user according to the matching degree of data and key word.

This preferred embodiment adopts the searching algorithm that catalogue retrieval and search engine combine, and can fast and accurately obtain Data.

Preferably, described unified monitoring management is carried out safely to each system, comprising:

(1) it is directed to data service system 40, data pretreatment 41, cloud storage encrypting and deciphering system 42, control system 43 Different security protections requires to take corresponding safety protection technique, is equipped with related safety protection equipment, forms complete peace Full protection system；

(2) set up effective Data Security, the safety in data storage, transmission, access process is carried out with comprehensive examining Consider, not only data is encrypted, the host-host protocol of data is encrypted simultaneously；

(3) set up virus and wooden horse defense mechanism, regularly update virus base and upgrading fire wall, the update cycle is t, and t takes It is worth for 6-10 days, the abnormal data detecting will be analyzed, and send early warning.

This preferred embodiment achieves the unified monitoring management to each system safety.

Preferably, described data pretreatment 41 includes data partitioning unit, data pick-up unit and access control peace Full policy optimization unit, described data partitioning unit is used for the speech data of need secrecy is divided into the data set of multiple mutual exclusions Close；Described data pick-up unit is used for the data acquisition system of described mutual exclusion is ranked up according to self-defining ordering rule, will be every First data cell in individual data acquisition system sequentially extracts, and preserves as small block data together with described ordering rule, There is not any association between the data cell two-by-two that wherein said mutual exclusion represents in data acquisition system；Described access control safety plan Slightly optimize the access control that unit is used for the access control safety policy optimization method generation system based on fine granularity division of resources Security strategy, comprising:

(1) data acquisition system based on the mutual exclusion after data pick-up cell processing, builds hierarchical data table structure, described Hierarchical data tree construction is three layer data tree constructions, and it includes service layer, logical layer and physical layer, and described service layer is and number According to the related root vertex of dispatch service, described logical layer is the data of association in access control safety strategy, described physical layer Comprise the data cell in the data acquisition system of all mutual exclusions；

(2) access control safety of the data for different safety class is formulated based on access control markup language xacml Strategy, the rule with data association in access control safety strategy is projected to the data cell in the data acquisition system of described mutual exclusion On, thus the rule in access control safety strategy is refine to data dimension；

(3) the enterprising line discipline of the data cell in the data acquisition system of each described mutual exclusion optimization, to delete distribution every The conflict of the rule in individual data cell and redundancy；

(4) merge the rule after optimizing, generate the access control safety strategy optimizing.

Preferably, described by data storage to corresponding storage device, comprising:

(1) small block data is stored to local storage, and using user-defined encryption technology, small block data is carried out Encryption；

(2) store in described cloud storage module 3 after remaining data being encrypted by cloud storage encrypting and deciphering system 42；Its In, after cloud storage module 3 receives data, cloud carries out to this data being saved in memory node after completeness check.

Above-mentioned two preferred embodiment arranges data pretreatment 41, first carries out data to the speech data of need secrecy and divides Cut data extraction to process, then the rule refinement controlling in security strategy that conducts interviews, it is possible to reduce the physics of data storage is deposited Storage space, reduces the expense of storage, and eliminates the conflict in access control safety strategy and redundancy, improves access control decision effect Rate；Extraction partial data is processed by data pick-up and stores in local storage, remainder data arranges corresponding access control Store after security strategy to cloud storage module 3, solve traditional cloud storage data-privacy based on simple encryption technology and protect The larger overhead of ratio that barrier mechanism is brought in actual process operation data and loaded down with trivial details, can effectively prevent malicious user Or cloud storage manager illegally steals, distorts the private data of user, improve the safety of the speech data storage that need to maintain secrecy Energy.

Preferably, described cloud storage encrypting and deciphering system 42 main by data owner, attribute mechanism, cloud, credible tripartite, use Five, family entity is constituted, and the described speech data to need secrecy is encrypted or deciphers, comprising:

(1) credible tripartite is user and attribute mechanism is respectively allocated User Identity uaid and attribute authority identity mark Aid, comprising:

A, initialized, credible tripartite's initialization system parameter isWherein α is random integers；

B, for each validated user, credible tripartite distribution uaid simultaneously Generates Certificate for it:

$certificate\left(uaid\right)=\hat{e}{\left(h\right(uaid),g)}^{c_{uaid}}$

Meanwhile, announce the authentication parameter of validated userWherein, c_uaid∈z_p；

C, generate identity key pair for data owner and validated user；

(2) generate the encryption and decryption key of identity-based, attribute encryption and decryption key and act on behalf of re-encrypted private key, wherein said The encryption and decryption key of identity-based includes identity public key gk_uaidWith identity private key ck_uaid, described attribute encryption and decryption key include belong to Property public key gk_aidWith attribute private key ck_aid:

${\mathrm{gk}}_{uaid}=\hat{e}{(g,g)}^{{\&proportional;}_{aid}}$

${\mathrm{gk}}_{aid}=\{\∀x\&element;{\mathrm{as}}_{aid}:{\mathrm{gk}}_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}}\}$

ck_uaid=(∝_aid,β_aid)

${\mathrm{ck}}_{aid}=(k_0=g^{{\&proportional;}_{aid}}{g}^{\mathrm{\α}\mathrm{\γ}},k_1=g^{\mathrm{\α}\mathrm{\γ}},\∀x\&element;{\mathrm{as}}_{uaid,aid}:k_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}\mathrm{\γ}})$

Wherein, as_aidThe community set that can distribute for single attribute mechanism, gk_xFor the public key of attribute x, b_xFor attribute x's Version number, ∝_aidFor the private key parameter of attribute mechanism, β_aidFor attribute undated parameter, as_uaid,aidIt is the identity according to attribute mechanism The community set of distribution, the parameter that γ randomly chooses for attribute mechanism, γ, ∝_aid,β_aid∈z_p；

(3) cloud storage encrypting and deciphering system 42 carries out data encryption using data key to speech data, obtains ciphertext ct, so After be utilized respectively identity public key and attribute public key to data key encrypt, generate identity key ciphertext ct_uWith attribute key ciphertext ct_a, comprising:

Character string ik of a, at random two regular lengths of generation, ak, merge and generate data key dk:

Dk=ik | | ak

B, carry out data encryption using the data that data key dk stores cloud storage module 3 to need, after obtaining ciphertext ct, Using attribute public key, ak is encrypted, generate attribute key ciphertext ct_a, using identity public key, ik is encrypted, generates identity key close Civilian ct_u；

(4) carry out acting on behalf of re-encryption, when receiving the request of data of user, cloud is close by identity using acting on behalf of re-encrypted private key Key ciphertext ct_uIt is converted into the ciphertext that specified user can decipher, wherein said re-encrypted private key of acting on behalf of uses itself by data owner Private key and identity public key calculate and generate；

(5), when carrying out data deciphering, after user receives data, it is utilized respectively identity private key ck_uaidWith attribute private key ck_aid Decryption identity key ciphertext ct_uWith attribute key ciphertext ct_a, then reconstruct data key, decrypting ciphertext ct；

(6) carry out the renewal of attribute and identity key.

This preferred embodiment passes through to arrange cloud storage encrypting and deciphering system 42, is capable of the fine granularity to eurypalynous data Access control and secret protection, resist user and the collusion of attribute mechanism simultaneously；Speech data to need secrecy, constructs respectively and is based on The encryption and decryption key of identity, attribute encryption and decryption key, merge composition data encryption key and this data are encrypted, thus only Meet identity simultaneously and the user of attribute double condition can decipher, greatly improve the safety of data safety management system 4 Energy.

In this application scenarios, update cycle t takes 7, and the safety of system improves 11% relatively.

Application scenarios 3

Referring to Fig. 1, Fig. 2, the speech recognition system based on cloud computing of an embodiment of this application scene, including voice Input module 1, load balancing control module 2, speech recognition platforms 3, speech cloud data base 5, voice input module 1 is with load all Weighing apparatus control module 2 connects, and load balancing control module 2 is connected with speech recognition platforms 3, speech recognition platforms 3 and speech cloud number Connect according to storehouse 5；Described voice input module 1 is used for input needs the speech data of identification, and speech data is sent to described Load balancing control module 2；Described speech recognition platforms 3 include multiple speech recognition servers, and described load balancing controls mould Block 2 is for the load according to speech recognition server by voice recognition tasks dynamically distributes to idle speech recognition server It is identified；Described speech cloud data base 5 is used for storing the voice library template of HMM, and speech recognition server will The speech data needing identification is quickly compared with the data in speech cloud data base 5, is that user returns identification knot in time Really.、

Preferably, described speech recognition platforms 3 include at least two voices knowledges that can separately provide speech identifying function Other server.

The above embodiment of the present invention can the discrimination that exists of the speech recognition system in effectively solving correlation technique be low, storage The problems such as slow with the access response that computing resource anxiety causes, not only increase contrast run time, and the suitability is wide, thus Solve above-mentioned technical problem.

Preferably, connected with the pattern that cloud net connects between speech recognition server.

Connected with the pattern that cloud net connects between this preferred embodiment speech recognition server, improve speech-recognition services The access response speed of device.

Preferably, the described speech recognition system based on cloud computing is also included for carrying out safety management to speech data Data safety management system 4；Described data safety management system 4 includes data service system 40, data pretreatment 41, cloud Storage encrypting and deciphering system 42, control system 43 and security management center 44；Described data service system 40 is used for being responsible for voice number According to storage management, backup and inquire about；The speech data that described data pretreatment 41 is used for need maintain secrecy carries out pretreatment； Described cloud storage encrypting and deciphering system 42 is used for the speech data that need maintain secrecy being carried out add according to the access control safety strategy optimizing Close or deciphering；Described control system 43 is used for storing speech data to corresponding storage device；Described security management center 44 For unified monitoring management is carried out safely to each system.

This preferred embodiment constructs the system structure of data safety management system 4.

Preferably, storage, backup and the inquiry of described responsible speech data, comprising:

(1) data form is changed, set up and be applied to the form that non-relational database is stored；

(2) basic data and expert data are splitted data into, the strategy using centralized and distributed combination enters to data Row storage, during storage, all data are all backed up；The strategy inclusion of described centralized and distributed combination: for higher than default The basic data of frequency adopts centralised storage, adopts distributed storage for the expert data less than predeterminated frequency；

(3) set up corresponding data retrievad algorithm, data is carried out with quick-searching, described data retrievad algorithm adopts catalogue The mode that retrieval and search engine combine is carried out, and specifically includes: set up data directory, according to catalogue, data is tentatively examined Rope；Input key word in search engine, precise search is carried out to data；Search engine finds the number of coupling according to certain mode According to, and be ranked up feeding back to user according to the matching degree of data and key word.

This preferred embodiment adopts the searching algorithm that catalogue retrieval and search engine combine, and can fast and accurately obtain Data.

Preferably, described unified monitoring management is carried out safely to each system, comprising:

(1) it is directed to data service system 40, data pretreatment 41, cloud storage encrypting and deciphering system 42, control system 43 Different security protections requires to take corresponding safety protection technique, is equipped with related safety protection equipment, forms complete peace Full protection system；

(2) set up effective Data Security, the safety in data storage, transmission, access process is carried out with comprehensive examining Consider, not only data is encrypted, the host-host protocol of data is encrypted simultaneously；

(3) set up virus and wooden horse defense mechanism, regularly update virus base and upgrading fire wall, the update cycle is t, and t takes It is worth for 6-10 days, the abnormal data detecting will be analyzed, and send early warning.

This preferred embodiment achieves the unified monitoring management to each system safety.

Preferably, described data pretreatment 41 includes data partitioning unit, data pick-up unit and access control peace Full policy optimization unit, described data partitioning unit is used for the speech data of need secrecy is divided into the data set of multiple mutual exclusions Close；Described data pick-up unit is used for the data acquisition system of described mutual exclusion is ranked up according to self-defining ordering rule, will be every First data cell in individual data acquisition system sequentially extracts, and preserves as small block data together with described ordering rule, There is not any association between the data cell two-by-two that wherein said mutual exclusion represents in data acquisition system；Described access control safety plan Slightly optimize the access control that unit is used for the access control safety policy optimization method generation system based on fine granularity division of resources Security strategy, comprising:

(1) data acquisition system based on the mutual exclusion after data pick-up cell processing, builds hierarchical data table structure, described Hierarchical data tree construction is three layer data tree constructions, and it includes service layer, logical layer and physical layer, and described service layer is and number According to the related root vertex of dispatch service, described logical layer is the data of association in access control safety strategy, described physical layer Comprise the data cell in the data acquisition system of all mutual exclusions；

(2) access control safety of the data for different safety class is formulated based on access control markup language xacml Strategy, the rule with data association in access control safety strategy is projected to the data cell in the data acquisition system of described mutual exclusion On, thus the rule in access control safety strategy is refine to data dimension；

(3) the enterprising line discipline of the data cell in the data acquisition system of each described mutual exclusion optimization, to delete distribution every The conflict of the rule in individual data cell and redundancy；

(4) merge the rule after optimizing, generate the access control safety strategy optimizing.

Preferably, described by data storage to corresponding storage device, comprising:

(1) small block data is stored to local storage, and using user-defined encryption technology, small block data is carried out Encryption；

(2) store in described cloud storage module 3 after remaining data being encrypted by cloud storage encrypting and deciphering system 42；Its In, after cloud storage module 3 receives data, cloud carries out to this data being saved in memory node after completeness check.

Above-mentioned two preferred embodiment arranges data pretreatment 41, first carries out data to the speech data of need secrecy and divides Cut data extraction to process, then the rule refinement controlling in security strategy that conducts interviews, it is possible to reduce the physics of data storage is deposited Storage space, reduces the expense of storage, and eliminates the conflict in access control safety strategy and redundancy, improves access control decision effect Rate；Extraction partial data is processed by data pick-up and stores in local storage, remainder data arranges corresponding access control Store after security strategy to cloud storage module 3, solve traditional cloud storage data-privacy based on simple encryption technology and protect The larger overhead of ratio that barrier mechanism is brought in actual process operation data and loaded down with trivial details, can effectively prevent malicious user Or cloud storage manager illegally steals, distorts the private data of user, improve the safety of the speech data storage that need to maintain secrecy Energy.

Preferably, described cloud storage encrypting and deciphering system 42 main by data owner, attribute mechanism, cloud, credible tripartite, use Five, family entity is constituted, and the described speech data to need secrecy is encrypted or deciphers, comprising:

(1) credible tripartite is user and attribute mechanism is respectively allocated User Identity uaid and attribute authority identity mark Aid, comprising:

A, initialized, credible tripartite's initialization system parameter isWherein α is random integers；

B, for each validated user, credible tripartite distribution uaid simultaneously Generates Certificate for it:

$certificate\left(uaid\right)=\hat{e}{\left(h\right(uaid),g)}^{c_{uaid}}$

Meanwhile, announce the authentication parameter of validated userWherein, c_uaid∈z_p；

C, generate identity key pair for data owner and validated user；

(2) generate the encryption and decryption key of identity-based, attribute encryption and decryption key and act on behalf of re-encrypted private key, wherein said The encryption and decryption key of identity-based includes identity public key gk_uaidWith identity private key ck_uaid, described attribute encryption and decryption key include belong to Property public key gk_aidWith attribute private key ck_aid:

${\mathrm{gk}}_{uaid}=\hat{e}{(g,g)}^{{\&proportional;}_{aid}}$

${\mathrm{gk}}_{aid}=\{\∀x\&element;{\mathrm{as}}_{aid}:{\mathrm{gk}}_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}}\}$

ck_uaid=(∝_aid,β_aid)

${\mathrm{ck}}_{aid}=(k_0=g^{{\&proportional;}_{aid}}{g}^{\mathrm{\α}\mathrm{\γ}},k_1=g^{\mathrm{\α}\mathrm{\γ}},\∀x\&element;{\mathrm{as}}_{uaid,aid}:k_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}\mathrm{\γ}})$

Wherein, as_aidThe community set that can distribute for single attribute mechanism, gk_xFor the public key of attribute x, b_xFor attribute x's Version number, ∝_aidFor the private key parameter of attribute mechanism, β_aidFor attribute undated parameter, as_uaid,aidIt is the identity according to attribute mechanism The community set of distribution, the parameter that γ randomly chooses for attribute mechanism, γ, ∝_aid,β_aid∈z_p；

(3) cloud storage encrypting and deciphering system 42 carries out data encryption using data key to speech data, obtains ciphertext ct, so After be utilized respectively identity public key and attribute public key to data key encrypt, generate identity key ciphertext ct_uWith attribute key ciphertext ct_a, comprising:

Character string ik of a, at random two regular lengths of generation, ak, merge and generate data key dk:

Dk=ik | | ak

B, carry out data encryption using the data that data key dk stores cloud storage module 3 to need, after obtaining ciphertext ct, Using attribute public key, ak is encrypted, generate attribute key ciphertext ct_a, using identity public key, ik is encrypted, generates identity key close Civilian ct_u；

(4) carry out acting on behalf of re-encryption, when receiving the request of data of user, cloud is close by identity using acting on behalf of re-encrypted private key Key ciphertext ct_uIt is converted into the ciphertext that specified user can decipher, wherein said re-encrypted private key of acting on behalf of uses itself by data owner Private key and identity public key calculate and generate；

(5), when carrying out data deciphering, after user receives data, it is utilized respectively identity private key ck_uaidWith attribute private key ck_aid Decryption identity key ciphertext ct_uWith attribute key ciphertext ct_a, then reconstruct data key, decrypting ciphertext ct；

(6) carry out the renewal of attribute and identity key.

This preferred embodiment passes through to arrange cloud storage encrypting and deciphering system 42, is capable of the fine granularity to eurypalynous data Access control and secret protection, resist user and the collusion of attribute mechanism simultaneously；Speech data to need secrecy, constructs respectively and is based on The encryption and decryption key of identity, attribute encryption and decryption key, merge composition data encryption key and this data are encrypted, thus only Meet identity simultaneously and the user of attribute double condition can decipher, greatly improve the safety of data safety management system 4 Energy.

In this application scenarios, update cycle t takes 8, and the safety of system improves 10% relatively.

Application scenarios 4

Referring to Fig. 1, Fig. 2, the speech recognition system based on cloud computing of an embodiment of this application scene, including voice Input module 1, load balancing control module 2, speech recognition platforms 3, speech cloud data base 5, voice input module 1 is with load all Weighing apparatus control module 2 connects, and load balancing control module 2 is connected with speech recognition platforms 3, speech recognition platforms 3 and speech cloud number Connect according to storehouse 5；Described voice input module 1 is used for input needs the speech data of identification, and speech data is sent to described Load balancing control module 2；Described speech recognition platforms 3 include multiple speech recognition servers, and described load balancing controls mould Block 2 is for the load according to speech recognition server by voice recognition tasks dynamically distributes to idle speech recognition server It is identified；Described speech cloud data base 5 is used for storing the voice library template of HMM, and speech recognition server will The speech data needing identification is quickly compared with the data in speech cloud data base 5, is that user returns identification knot in time Really.、

Preferably, described speech recognition platforms 3 include at least two voices knowledges that can separately provide speech identifying function Other server.

The above embodiment of the present invention can the discrimination that exists of the speech recognition system in effectively solving correlation technique be low, storage The problems such as slow with the access response that computing resource anxiety causes, not only increase contrast run time, and the suitability is wide, thus Solve above-mentioned technical problem.

Preferably, connected with the pattern that cloud net connects between speech recognition server.

Connected with the pattern that cloud net connects between this preferred embodiment speech recognition server, improve speech-recognition services The access response speed of device.

Preferably, the described speech recognition system based on cloud computing is also included for carrying out safety management to speech data Data safety management system 4；Described data safety management system 4 includes data service system 40, data pretreatment 41, cloud Storage encrypting and deciphering system 42, control system 43 and security management center 44；Described data service system 40 is used for being responsible for voice number According to storage management, backup and inquire about；The speech data that described data pretreatment 41 is used for need maintain secrecy carries out pretreatment； Described cloud storage encrypting and deciphering system 42 is used for the speech data that need maintain secrecy being carried out add according to the access control safety strategy optimizing Close or deciphering；Described control system 43 is used for storing speech data to corresponding storage device；Described security management center 44 For unified monitoring management is carried out safely to each system.

This preferred embodiment constructs the system structure of data safety management system 4.

Preferably, storage, backup and the inquiry of described responsible speech data, comprising:

(1) data form is changed, set up and be applied to the form that non-relational database is stored；

(2) basic data and expert data are splitted data into, the strategy using centralized and distributed combination enters to data Row storage, during storage, all data are all backed up；The strategy inclusion of described centralized and distributed combination: for higher than default The basic data of frequency adopts centralised storage, adopts distributed storage for the expert data less than predeterminated frequency；

(3) set up corresponding data retrievad algorithm, data is carried out with quick-searching, described data retrievad algorithm adopts catalogue The mode that retrieval and search engine combine is carried out, and specifically includes: set up data directory, according to catalogue, data is tentatively examined Rope；Input key word in search engine, precise search is carried out to data；Search engine finds the number of coupling according to certain mode According to, and be ranked up feeding back to user according to the matching degree of data and key word.

This preferred embodiment adopts the searching algorithm that catalogue retrieval and search engine combine, and can fast and accurately obtain Data.

Preferably, described unified monitoring management is carried out safely to each system, comprising:

(1) it is directed to data service system 40, data pretreatment 41, cloud storage encrypting and deciphering system 42, control system 43 Different security protections requires to take corresponding safety protection technique, is equipped with related safety protection equipment, forms complete peace Full protection system；

(2) set up effective Data Security, the safety in data storage, transmission, access process is carried out with comprehensive examining Consider, not only data is encrypted, the host-host protocol of data is encrypted simultaneously；

(3) set up virus and wooden horse defense mechanism, regularly update virus base and upgrading fire wall, the update cycle is t, and t takes It is worth for 6-10 days, the abnormal data detecting will be analyzed, and send early warning.

This preferred embodiment achieves the unified monitoring management to each system safety.

Preferably, described data pretreatment 41 includes data partitioning unit, data pick-up unit and access control peace Full policy optimization unit, described data partitioning unit is used for the speech data of need secrecy is divided into the data set of multiple mutual exclusions Close；Described data pick-up unit is used for the data acquisition system of described mutual exclusion is ranked up according to self-defining ordering rule, will be every First data cell in individual data acquisition system sequentially extracts, and preserves as small block data together with described ordering rule, There is not any association between the data cell two-by-two that wherein said mutual exclusion represents in data acquisition system；Described access control safety plan Slightly optimize the access control that unit is used for the access control safety policy optimization method generation system based on fine granularity division of resources Security strategy, comprising:

(1) data acquisition system based on the mutual exclusion after data pick-up cell processing, builds hierarchical data table structure, described Hierarchical data tree construction is three layer data tree constructions, and it includes service layer, logical layer and physical layer, and described service layer is and number According to the related root vertex of dispatch service, described logical layer is the data of association in access control safety strategy, described physical layer Comprise the data cell in the data acquisition system of all mutual exclusions；

(2) access control safety of the data for different safety class is formulated based on access control markup language xacml Strategy, the rule with data association in access control safety strategy is projected to the data cell in the data acquisition system of described mutual exclusion On, thus the rule in access control safety strategy is refine to data dimension；

(3) the enterprising line discipline of the data cell in the data acquisition system of each described mutual exclusion optimization, to delete distribution every The conflict of the rule in individual data cell and redundancy；

(4) merge the rule after optimizing, generate the access control safety strategy optimizing.

Preferably, described by data storage to corresponding storage device, comprising:

(1) small block data is stored to local storage, and using user-defined encryption technology, small block data is carried out Encryption；

(2) store in described cloud storage module 3 after remaining data being encrypted by cloud storage encrypting and deciphering system 42；Its In, after cloud storage module 3 receives data, cloud carries out to this data being saved in memory node after completeness check.

Above-mentioned two preferred embodiment arranges data pretreatment 41, first carries out data to the speech data of need secrecy and divides Cut data extraction to process, then the rule refinement controlling in security strategy that conducts interviews, it is possible to reduce the physics of data storage is deposited Storage space, reduces the expense of storage, and eliminates the conflict in access control safety strategy and redundancy, improves access control decision effect Rate；Extraction partial data is processed by data pick-up and stores in local storage, remainder data arranges corresponding access control Store after security strategy to cloud storage module 3, solve traditional cloud storage data-privacy based on simple encryption technology and protect The larger overhead of ratio that barrier mechanism is brought in actual process operation data and loaded down with trivial details, can effectively prevent malicious user Or cloud storage manager illegally steals, distorts the private data of user, improve the safety of the speech data storage that need to maintain secrecy Energy.

Preferably, described cloud storage encrypting and deciphering system 42 main by data owner, attribute mechanism, cloud, credible tripartite, use Five, family entity is constituted, and the described speech data to need secrecy is encrypted or deciphers, comprising:

(1) credible tripartite is user and attribute mechanism is respectively allocated User Identity uaid and attribute authority identity mark Aid, comprising:

A, initialized, credible tripartite's initialization system parameter isWherein α is random integers；

B, for each validated user, credible tripartite distribution uaid simultaneously Generates Certificate for it:

$certificate\left(uaid\right)=\hat{e}{\left(h\right(uaid),g)}^{c_{uaid}}$

Meanwhile, announce the authentication parameter of validated userWherein, c_uaid∈z_p；

C, generate identity key pair for data owner and validated user；

(2) generate the encryption and decryption key of identity-based, attribute encryption and decryption key and act on behalf of re-encrypted private key, wherein said The encryption and decryption key of identity-based includes identity public key gk_uaidWith identity private key ck_uaid, described attribute encryption and decryption key include belong to Property public key gk_aidWith attribute private key ck_aid:

${\mathrm{gk}}_{uaid}=\hat{e}{(g,g)}^{{\&proportional;}_{aid}}$

${\mathrm{gk}}_{aid}=\{\∀x\&element;{\mathrm{as}}_{aid}:{\mathrm{gk}}_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}}\}$

ck_uaid=(∝_aid,β_aid)

${\mathrm{ck}}_{aid}=(k_0=g^{{\&proportional;}_{aid}}{g}^{\mathrm{\α}\mathrm{\γ}},k_1=g^{\mathrm{\α}\mathrm{\γ}},\∀x\&element;{\mathrm{as}}_{uaid,aid}:k_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}\mathrm{\γ}})$

Wherein, as_aidThe community set that can distribute for single attribute mechanism, gk_xFor the public key of attribute x, b_xFor attribute x's Version number, ∝_aidFor the private key parameter of attribute mechanism, β_aidFor attribute undated parameter, as_uaid,aidIt is the identity according to attribute mechanism The community set of distribution, the parameter that γ randomly chooses for attribute mechanism, γ, ∝_aid,β_aid∈z_p；

(3) cloud storage encrypting and deciphering system 42 carries out data encryption using data key to speech data, obtains ciphertext ct, so After be utilized respectively identity public key and attribute public key to data key encrypt, generate identity key ciphertext ct_uWith attribute key ciphertext ct_a, comprising:

Character string ik of a, at random two regular lengths of generation, ak, merge and generate data key dk:

Dk=ik | | ak

B, carry out data encryption using the data that data key dk stores cloud storage module 3 to need, after obtaining ciphertext ct, Using attribute public key, ak is encrypted, generate attribute key ciphertext ct_a, using identity public key, ik is encrypted, generates identity key close Civilian ct_u；

(4) carry out acting on behalf of re-encryption, when receiving the request of data of user, cloud is close by identity using acting on behalf of re-encrypted private key Key ciphertext ct_uIt is converted into the ciphertext that specified user can decipher, wherein said re-encrypted private key of acting on behalf of uses itself by data owner Private key and identity public key calculate and generate；

(5), when carrying out data deciphering, after user receives data, it is utilized respectively identity private key ck_uaidWith attribute private key ck_aid Decryption identity key ciphertext ct_uWith attribute key ciphertext ct_a, then reconstruct data key, decrypting ciphertext ct；

(6) carry out the renewal of attribute and identity key.

This preferred embodiment passes through to arrange cloud storage encrypting and deciphering system 42, is capable of the fine granularity to eurypalynous data Access control and secret protection, resist user and the collusion of attribute mechanism simultaneously；Speech data to need secrecy, constructs respectively and is based on The encryption and decryption key of identity, attribute encryption and decryption key, merge composition data encryption key and this data are encrypted, thus only Meet identity simultaneously and the user of attribute double condition can decipher, greatly improve the safety of data safety management system 4 Energy.

In this application scenarios, update cycle t takes 9, and the safety of system improves 9% relatively.

Application scenarios 5

Referring to Fig. 1, Fig. 2, the speech recognition system based on cloud computing of an embodiment of this application scene, including voice Input module 1, load balancing control module 2, speech recognition platforms 3, speech cloud data base 5, voice input module 1 is with load all Weighing apparatus control module 2 connects, and load balancing control module 2 is connected with speech recognition platforms 3, speech recognition platforms 3 and speech cloud number Connect according to storehouse 5；Described voice input module 1 is used for input needs the speech data of identification, and speech data is sent to described Load balancing control module 2；Described speech recognition platforms 3 include multiple speech recognition servers, and described load balancing controls mould Block 2 is for the load according to speech recognition server by voice recognition tasks dynamically distributes to idle speech recognition server It is identified；Described speech cloud data base 5 is used for storing the voice library template of HMM, and speech recognition server will The speech data needing identification is quickly compared with the data in speech cloud data base 5, is that user returns identification knot in time Really.、

Preferably, described speech recognition platforms 3 include at least two voices knowledges that can separately provide speech identifying function Other server.

The above embodiment of the present invention can the discrimination that exists of the speech recognition system in effectively solving correlation technique be low, storage The problems such as slow with the access response that computing resource anxiety causes, not only increase contrast run time, and the suitability is wide, thus Solve above-mentioned technical problem.

Preferably, connected with the pattern that cloud net connects between speech recognition server.

Connected with the pattern that cloud net connects between this preferred embodiment speech recognition server, improve speech-recognition services The access response speed of device.

Preferably, the described speech recognition system based on cloud computing is also included for carrying out safety management to speech data Data safety management system 4；Described data safety management system 4 includes data service system 40, data pretreatment 41, cloud Storage encrypting and deciphering system 42, control system 43 and security management center 44；Described data service system 40 is used for being responsible for voice number According to storage management, backup and inquire about；The speech data that described data pretreatment 41 is used for need maintain secrecy carries out pretreatment； Described cloud storage encrypting and deciphering system 42 is used for the speech data that need maintain secrecy being carried out add according to the access control safety strategy optimizing Close or deciphering；Described control system 43 is used for storing speech data to corresponding storage device；Described security management center 44 For unified monitoring management is carried out safely to each system.

This preferred embodiment constructs the system structure of data safety management system 4.

Preferably, storage, backup and the inquiry of described responsible speech data, comprising:

(1) data form is changed, set up and be applied to the form that non-relational database is stored；

(2) basic data and expert data are splitted data into, the strategy using centralized and distributed combination enters to data Row storage, during storage, all data are all backed up；The strategy inclusion of described centralized and distributed combination: for higher than default The basic data of frequency adopts centralised storage, adopts distributed storage for the expert data less than predeterminated frequency；

(3) set up corresponding data retrievad algorithm, data is carried out with quick-searching, described data retrievad algorithm adopts catalogue The mode that retrieval and search engine combine is carried out, and specifically includes: set up data directory, according to catalogue, data is tentatively examined Rope；Input key word in search engine, precise search is carried out to data；Search engine finds the number of coupling according to certain mode According to, and be ranked up feeding back to user according to the matching degree of data and key word.

This preferred embodiment adopts the searching algorithm that catalogue retrieval and search engine combine, and can fast and accurately obtain Data.

Preferably, described unified monitoring management is carried out safely to each system, comprising:

(1) it is directed to data service system 40, data pretreatment 41, cloud storage encrypting and deciphering system 42, control system 43 Different security protections requires to take corresponding safety protection technique, is equipped with related safety protection equipment, forms complete peace Full protection system；

(2) set up effective Data Security, the safety in data storage, transmission, access process is carried out with comprehensive examining Consider, not only data is encrypted, the host-host protocol of data is encrypted simultaneously；

(3) set up virus and wooden horse defense mechanism, regularly update virus base and upgrading fire wall, the update cycle is t, and t takes It is worth for 6-10 days, the abnormal data detecting will be analyzed, and send early warning.

This preferred embodiment achieves the unified monitoring management to each system safety.

Preferably, described data pretreatment 41 includes data partitioning unit, data pick-up unit and access control peace Full policy optimization unit, described data partitioning unit is used for the speech data of need secrecy is divided into the data set of multiple mutual exclusions Close；Described data pick-up unit is used for the data acquisition system of described mutual exclusion is ranked up according to self-defining ordering rule, will be every First data cell in individual data acquisition system sequentially extracts, and preserves as small block data together with described ordering rule, There is not any association between the data cell two-by-two that wherein said mutual exclusion represents in data acquisition system；Described access control safety plan Slightly optimize the access control that unit is used for the access control safety policy optimization method generation system based on fine granularity division of resources Security strategy, comprising:

(1) data acquisition system based on the mutual exclusion after data pick-up cell processing, builds hierarchical data table structure, described Hierarchical data tree construction is three layer data tree constructions, and it includes service layer, logical layer and physical layer, and described service layer is and number According to the related root vertex of dispatch service, described logical layer is the data of association in access control safety strategy, described physical layer Comprise the data cell in the data acquisition system of all mutual exclusions；

(2) access control safety of the data for different safety class is formulated based on access control markup language xacml Strategy, the rule with data association in access control safety strategy is projected to the data cell in the data acquisition system of described mutual exclusion On, thus the rule in access control safety strategy is refine to data dimension；

(3) the enterprising line discipline of the data cell in the data acquisition system of each described mutual exclusion optimization, to delete distribution every The conflict of the rule in individual data cell and redundancy；

(4) merge the rule after optimizing, generate the access control safety strategy optimizing.

Preferably, described by data storage to corresponding storage device, comprising:

(1) small block data is stored to local storage, and using user-defined encryption technology, small block data is carried out Encryption；

(2) store in described cloud storage module 3 after remaining data being encrypted by cloud storage encrypting and deciphering system 42；Its In, after cloud storage module 3 receives data, cloud carries out to this data being saved in memory node after completeness check.

Above-mentioned two preferred embodiment arranges data pretreatment 41, first carries out data to the speech data of need secrecy and divides Cut data extraction to process, then the rule refinement controlling in security strategy that conducts interviews, it is possible to reduce the physics of data storage is deposited Storage space, reduces the expense of storage, and eliminates the conflict in access control safety strategy and redundancy, improves access control decision effect Rate；Extraction partial data is processed by data pick-up and stores in local storage, remainder data arranges corresponding access control Store after security strategy to cloud storage module 3, solve traditional cloud storage data-privacy based on simple encryption technology and protect The larger overhead of ratio that barrier mechanism is brought in actual process operation data and loaded down with trivial details, can effectively prevent malicious user Or cloud storage manager illegally steals, distorts the private data of user, improve the safety of the speech data storage that need to maintain secrecy Energy.

Preferably, described cloud storage encrypting and deciphering system 42 main by data owner, attribute mechanism, cloud, credible tripartite, use Five, family entity is constituted, and the described speech data to need secrecy is encrypted or deciphers, comprising:

(1) credible tripartite is user and attribute mechanism is respectively allocated User Identity uaid and attribute authority identity mark Aid, comprising:

A, initialized, credible tripartite's initialization system parameter isWherein α is random integers；

B, for each validated user, credible tripartite distribution uaid simultaneously Generates Certificate for it:

$certificate\left(uaid\right)=\hat{e}{\left(h\right(uaid),g)}^{c_{uaid}}$

Meanwhile, announce the authentication parameter of validated userWherein, c_uaid∈z_p；

C, generate identity key pair for data owner and validated user；

(2) generate the encryption and decryption key of identity-based, attribute encryption and decryption key and act on behalf of re-encrypted private key, wherein said The encryption and decryption key of identity-based includes identity public key gk_uaidWith identity private key ck_uaid, described attribute encryption and decryption key include belong to Property public key gk_aidWith attribute private key ck_aid:

${\mathrm{gk}}_{uaid}=\hat{e}{(g,g)}^{{\&proportional;}_{aid}}$

${\mathrm{gk}}_{aid}=\{\∀x\&element;{\mathrm{as}}_{aid}:{\mathrm{gk}}_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}}\}$

ck_uaid=(∝_aid,β_aid)

${\mathrm{ck}}_{aid}=(k_0=g^{{\&proportional;}_{aid}}{g}^{\mathrm{\α}\mathrm{\γ}},k_1=g^{\mathrm{\α}\mathrm{\γ}},\∀x\&element;{\mathrm{as}}_{uaid,aid}:k_x=h{\left(x\right)}^{b_x{\mathrm{\β}}_{aid}\mathrm{\γ}})$

Wherein, as_aidThe community set that can distribute for single attribute mechanism, gk_xFor the public key of attribute x, b_xFor attribute x's Version number, ∝_aidFor the private key parameter of attribute mechanism, β_aidFor attribute undated parameter, as_uaid,aidIt is the identity according to attribute mechanism The community set of distribution, the parameter that γ randomly chooses for attribute mechanism, γ, ∝_aid,β_aid∈z_p；

(3) cloud storage encrypting and deciphering system 42 carries out data encryption using data key to speech data, obtains ciphertext ct, so After be utilized respectively identity public key and attribute public key to data key encrypt, generate identity key ciphertext ct_uWith attribute key ciphertext ct_a, comprising:

Character string ik of a, at random two regular lengths of generation, ak, merge and generate data key dk:

Dk=ik | | ak

B, carry out data encryption using the data that data key dk stores cloud storage module 3 to need, after obtaining ciphertext ct, Using attribute public key, ak is encrypted, generate attribute key ciphertext ct_a, using identity public key, ik is encrypted, generates identity key close Civilian ct_u；

(4) carry out acting on behalf of re-encryption, when receiving the request of data of user, cloud is close by identity using acting on behalf of re-encrypted private key Key ciphertext ct_uIt is converted into the ciphertext that specified user can decipher, wherein said re-encrypted private key of acting on behalf of uses itself by data owner Private key and identity public key calculate and generate；

(5), when carrying out data deciphering, after user receives data, it is utilized respectively identity private key ck_uaidWith attribute private key ck_aid Decryption identity key ciphertext ct_uWith attribute key ciphertext ct_a, then reconstruct data key, decrypting ciphertext ct；

(6) carry out the renewal of attribute and identity key.

This preferred embodiment passes through to arrange cloud storage encrypting and deciphering system 42, is capable of the fine granularity to eurypalynous data Access control and secret protection, resist user and the collusion of attribute mechanism simultaneously；Speech data to need secrecy, constructs respectively and is based on The encryption and decryption key of identity, attribute encryption and decryption key, merge composition data encryption key and this data are encrypted, thus only Meet identity simultaneously and the user of attribute double condition can decipher, greatly improve the safety of data safety management system 4 Energy.

In this application scenarios, update cycle t takes 10, and the safety of system improves 8% relatively.

Finally it should be noted that above example is only in order to illustrating technical scheme, rather than the present invention is protected The restriction of shield scope, although having made to explain to the present invention with reference to preferred embodiment, those of ordinary skill in the art should Work as understanding, technical scheme can be modified or equivalent, without deviating from the reality of technical solution of the present invention Matter and scope.

Claims (3)

1. a kind of speech recognition system based on cloud computing is it is characterised in that include voice input module, load balancing controls mould Block, speech recognition platforms, speech cloud data base, voice input module is connected with load balancing control module, and load balancing controls Module is connected with speech recognition platforms, and speech recognition platforms are connected with speech cloud data base；Described voice input module is used for defeated Enter to need the speech data of identification, and speech data is sent to described load balancing control module；Described speech recognition platforms Including multiple speech recognition servers, described load balancing control module is used for load according to speech recognition server by voice Identification mission dynamically distributes are identified to idle speech recognition server；Described speech cloud data base is used for storing hidden horse The voice library template of Er Kefu model, the number in speech data and speech cloud data base that speech recognition server will need to identify According to quickly being compared, it is that user returns recognition result in time.

2. a kind of speech recognition system based on cloud computing according to claim 1 is it is characterised in that described speech recognition Platform includes at least two speech recognition servers that can separately provide speech identifying function.

3. a kind of speech recognition system based on cloud computing according to claim 2 is it is characterised in that speech-recognition services Connected with the pattern that cloud net connects between device.