阅读说明：本技术 基于声纹的验证方法及装置 (Verification method and device based on vocal print ) 是由 葛亮 于 2019-09-11 设计创作，主要内容包括：本发明公开了一种基于声纹的验证方法及装置。其中基于声纹的验证方法,包括步骤：采集至少一个用户的语音信息并提取各用户的语音特征存储至模型库中；随机生成验证密码供待验证用户朗读,并获取朗读语音；提取所述朗读语音中待验证的内容及语音特征,根据所述模型库以及验证密码对所述待验证的内容和语音特征进行验证,并得到验证结果。本发明是基于本地离线处理的声纹识别方法,每次的识别内容随机变化,并对验证语音信号进行活体检测,极大提升了用户验证的安全性。(The invention discloses a kind of verification method and device based on vocal print.Wherein based on the verification method of vocal print, store comprising steps of acquiring the voice messaging of at least one user and extracting the phonetic feature of each user into model library；The random verifying password that generates is read aloud for user to be verified, and is obtained and read aloud voice；Content and phonetic feature to be verified in voice are read aloud described in extraction, the content and phonetic feature to be verified are verified according to the model library and verifying password, and be verified result.The present invention is the method for recognizing sound-groove based on local processed offline, each identification content changes at random, and carries out In vivo detection, the significant increase safety of user's checking to verifying voice signal.)

1. a kind of verification method based on vocal print, which is characterized in that comprising steps of

Step 1, it acquires the voice messaging of at least one user and extracts the phonetic feature of each user and store into model library；

Step 2, the random verifying password that generates is read aloud for user to be verified, and is obtained and read aloud voice；

Step 3, content and phonetic feature to be verified in voice are read aloud described in extraction, according to the model library and verifying password The content and phonetic feature to be verified are verified, and are verified result.

2. as described in claim 1 based on the verification method of vocal print, which is characterized in that the phonetic feature for extracting each user It stores and is specifically included into model library:

N sections of sound bites, each sound bite is respectively cut into identical at least two voice messagings of single user's content Content be a character；

The phonetic feature for extracting each sound bite, by the phonetic feature of the identical sound bite of content in all voice messagings It compares, extracts the phonetic feature of user's corresponding content；

The sonification model that the phonetic feature of all different contents of each user is combined into the user is stored to model library In.

3. as claimed in claim 2 based on the verification method of vocal print, which is characterized in that fail when the sound bite is cut When, it reminds user and resurveys the voice messaging of user.

4. as described in claim 1 based on the verification method of vocal print, which is characterized in that the verifying password is by least one word Symbol composition；Character quantity for storing the dictionary of the character is less than or equal to the corresponding content of voice messaging of the user Character quantity.

5. as claimed in claim 2 based on the verification method of vocal print, which is characterized in that the step 3 specifically includes:

The voice of reading aloud is cut into m sections of sound bites to be verified, the content of each sound bite to be verified is a word Symbol, m≤n；

Extract the phonetic feature of each sound bite to be verified；

By the sound bite that user is corresponded in the model library according to it is described verifying password content be combined, and with combination after The sequence sound bite to be verified of reading aloud voice with described correspondingly carry out phonetic feature matching；

If being matched to corresponding user, successful match, otherwise it fails to match.

6. as claimed in claim 2 based on the verification method of vocal print, which is characterized in that the step 3 specifically includes:

Whether the content that voice is read aloud described in verifying is identical as the verifying content of password；

If they are the same, then the voice of reading aloud is cut into m sections of sound bites to be verified, the content of each sound bite to be verified For a character, m≤n；

Extract the phonetic feature of each sound bite to be verified；

The sound bite to be verified for reading aloud voice is subjected to voice with the speech model of the corresponding user in the model library Characteristic matching；

If being matched to corresponding user, successful match, otherwise it fails to match.

7. such as the verification method described in claim 5 or 6 based on vocal print, which is characterized in that when the sound bite to be verified When cutting failure, reminds user to be verified and resurvey the described of the user to be verified and read aloud voice.

8. as described in claim 1 based on the verification method of vocal print, which is characterized in that when the user to be verified is matched to pair When the user answered, according to the permission that the user is possessed, the corresponding permission of the user to be verified is assigned.

9. as claimed in claim 3 based on the verification method of vocal print, which is characterized in that the model library and dictionary are local Database.

10. a kind of device for having used the verification method as claimed in any one of claims 1 to 9 based on vocal print, comprising: use Language in the sound acquisition module for acquiring the voice messaging with reading aloud voice, for extracting the voice messaging He reading aloud voice The characteristic extracting module of sound feature is extracted the character at random and is generated for storing the memory module of the model library and character It is described verifying password secret generation module, to it is described read aloud voice and verify password carry out matching verifying authentication module.

Technical field

The present invention relates to identity validation technologies, such as financial payment, gate inhibition, electronic lock device etc., more particularly to are based on sound The identity validation technology of line, and it is loaded with the computer equipment and storage equipment of the identity validation technology.

Background technique

Application on Voiceprint Recognition has had been applied in many fields at present, such as when voice control, specified by allowing user to say Instruction, to execute corresponding movement for the specified instruction, i.e., the voice control is relevant based on content.Also have Speech recognition is by recording audio in advance, and identical user says the audio of identical content later, then is identified by, still This audio recorded in advance is easier to be cracked, and verification technique of some based on cloud, and needing first will be on voice Network is reached, may be subjected to the attack of hacker during this, therefore there are security risks for this voice recognition mode.

Summary of the invention

In order to solve the technical problem that voice print verification degree of safety is lower in the prior art, the invention proposes based on vocal print Verification method and device.

Verification method proposed by the present invention based on vocal print, comprising steps of

Step 1, it acquires the voice messaging of at least one user and extracts the phonetic feature of each user and store into model library；

Step 2, the random verifying password that generates is read aloud for user to be verified, and is obtained and read aloud voice；

Step 3, content and phonetic feature to be verified in voice are read aloud described in extraction, according to the model library and verifying password The content and phonetic feature to be verified are verified, and are verified result.

Specifically, the phonetic feature for extracting each user is stored and is specifically included into model library:

N sections of sound bites, each sound bite is respectively cut into identical at least two voice messagings of single user's content Content be a character；

The phonetic feature for extracting each sound bite, by the phonetic feature of the identical sound bite of content in all voice messagings It compares, extracts the phonetic feature of user's corresponding content；

The sonification model that the phonetic feature of all different contents of each user is combined into the user is stored to model library In.

Preferably, it when the sound bite cuts failure, reminds user and resurveys the voice messaging of user.

Preferably, the verifying password is made of at least one character.

Preferably, the voice messaging that the user is less than or equal to for storing the character quantity of the dictionary of the character corresponds to Content character quantity.

In one embodiment, the step 3 specifically includes: the voice of reading aloud is cut into m sections of voice sheets to be verified Section, the content of each sound bite to be verified are a character, m≤n；The voice for extracting each sound bite to be verified is special Sign；By the sound bite that user is corresponded in the model library according to it is described verifying password content be combined, and with combination after The sequence sound bite to be verified of reading aloud voice with described correspondingly carry out phonetic feature matching；If being matched to corresponding User, then successful match, otherwise it fails to match.

In another embodiment, the step 3 specifically includes: whether the content that voice is read aloud described in verifying is tested with described The content for demonstrate,proving password is identical；If they are the same, then the voice of reading aloud is cut into m sections of sound bites to be verified, each is to be verified The content of sound bite is a character, m≤n；Extract the phonetic feature of each sound bite to be verified；Language is read aloud by described The sound bite to be verified of sound carries out phonetic feature matching with the speech model of the corresponding user in the model library；If being matched to Corresponding user, then successful match, otherwise it fails to match.

Preferably, when the sound bite to be verified cutting failure, remind user to be verified and resurvey it is described to The described of verifying user reads aloud voice.

Preferably, it when the user to be verified is matched to corresponding user, according to the permission that the user is possessed, assigns Give the corresponding permission of the user to be verified.

Preferably, the model library and dictionary are local database.

The invention also provides the devices for having used the verification method described in above-mentioned technical proposal based on vocal print, comprising: Sound acquisition module for acquiring the voice messaging He reading aloud voice, for extracting the voice messaging and reading aloud voice It is raw to extract the character for storing the memory module of the model library and character at random for the characteristic extracting module of phonetic feature At it is described verifying password secret generation module, to it is described read aloud voice and verify password carry out matching verifying authentication module.

Computer equipment proposed by the present invention, including memory, processor and storage are on a memory and can be in processor The computer program of upper operation, the processor are realized described in above-mentioned technical proposal when executing the computer program based on sound The verification method of line.

Storage equipment proposed by the present invention comprising computer executable instructions, described includes computer executable instructions Storage equipment realized when executing the computer executable instructions as computer processor described in above-mentioned technical proposal based on The verification method of vocal print.

The present invention collects the voice signal of user by sound acquisition modules such as microphones, and only in processing locality, no It can reveal user's personal information, and and Network Isolation, the safety of system has been significantly greatly increased.The present invention is believed by the vocal print of user Breath, identifies the exclusive biological characteristic of user, can greatly guarantee the exclusive permission of user, have very in the occasion of authentication Big application value.The present invention is based on the methods of limited dictionary, improve technical feasibility and user experience, enhance practical Property, it can guarantee the safety of verifying under finite computational abilities, and prevent information leakage with randomness content, ensure safety Under the premise of keep authentication safer.

Detailed description of the invention

Below with reference to embodiment and attached drawing, the present invention is described in detail, in which:

Fig. 1 is overall flow figure of the present invention.

Fig. 2 is register flow path figure of the invention.

Fig. 3 is verifying flow chart of the invention.

Specific embodiment

The principle and embodiment that the invention will now be described in detail with reference to the accompanying drawings.

It as shown in Figure 1 to Figure 3, include two parts of registration and verifying the present invention is based on the verification method of vocal print.

In registration, the present invention acquires the voice messaging of user and extracts the phonetic feature of user and stores into model library, Specific single user needs user to carry out model training to saving your breath twice as requested, extracts the use when registering The biological characteristic at family establishes the sonification model of user and is stored in local model library.The present invention can specifically pass through following step Suddenly model buildings are carried out to one or more users.

(a) verifying device is waken up, then starts to register；

(b) it records, records at least twice to active user, obtain at least two voice messagings as input signal；Such as It is required that user reads aloud twice of number of 0-9, two corresponding voice messagings are then generated.

(c) pretreatment such as noise reduction, cutting is carried out to the voice messaging of recording, voice messaging is cut into sound bite, often The content of one sound bite is a character, and character mentioned here includes at least one in number, letter, word and symbol Kind.In the present embodiment, every voice messaging is divided into the sound bite of 10 independent numbers.When sound bite cuts failure, User is reminded to record again then so as to resurvey the voice messaging of user, such as the number of segment of cutting is undesirable, then It needs again to record to user；

(d) feature extraction then is carried out to every voice messaging, the phonetic feature of each sound bite is extracted, by all voices The phonetic feature of the identical sound bite of content compares in information, extracts the phonetic feature of user's corresponding content, this reality It applies in example, the phonetic feature of the identical sound bite of two numbers is compared to extract the voice of user's corresponding content spy Sign, for example, two sections of sound bites of 0 this pronunciation are compared to extract phonetic feature when user issues 0.

(e) phonetic feature of all different contents of each user is combined into the sonification model of this user, and deposited It stores up in local model library.

In above-mentioned registration process, it is not limited to the numeric utterance of typing user, the word of user can also be employed Pronunciation, the pronunciation of word, the pronunciation of symbol etc., the character provided be also not limited to only be provided once, for example, passing through big data Analysis, some pronunciations are compared when being difficult to compare feature, reading can also be required in a voice messaging several times, to be conducive to Extract the phonetic feature that user is directed to the pronunciation.When relatively, the present invention is the sound bite of identical content to be carried out pair Than, therefore, the sound bite of even two voice messagings, it is also possible to occur being more than the identical voice sheet of 2 or more contents Section, this is also allowed.Certainly in general, a content compares twice can.Below with the 4 of family of four It carries out the explanation of registration one of example, needs each user to say two section 0 to 9 of digital speech under prompt, to every section of voice It is pre-processed, including noise reduction and voice are cut, then carry out MFCC feature extraction, save as two groups of phonetic features of active user Data, then phase knowledge and magnanimity calculating is carried out with two groups of voice feature datas, find out the sounding mould that suitable threshold value is saved in active user Type is used as verify data, is stored in model library.

User completes registration process, so that it may the verification process after carrying out, to realize desired control.The present invention Verification method there is randomness, be not to extract preprepared data out from limited dictionary to be matched. The present invention is that at least one character is randomly selected from dictionary, forms verifying password into row stochastic permutation and combination, then right again The content and vocal print of the voice of user to be verified carry out matching verifying, belong to the relevant verification method of half text of one kind.It takes out every time The quantity of the character taken can be fixed, and be also possible to unfixed, and those skilled in that art set i.e. as needed It can.

Verification process of the invention is also required to voice signal input, after user says one section of voice according to prompt, according to mentioning Whether the sonification model of the user of preceding foundation, matches voice signal and is calculated, obtain user to be verified in user model In library, specifically who user, specific verification process are as follows.

(a) firstly, waking up verifying device；

(b) device enters challenge-response mode, generates verifying password at random from dictionary and reads aloud for user to be verified, and to Voice is read aloud in the acquisition that record of reading aloud at family；Verifying password is made of at least one character, preferably, is verified close The character quantity of code is more, and sounding information obtained by device is more, and judgement in this way is also just more accurate, certainly, verifies close The character quantity of code is excessive, and will cause the burden of device, and therefore, general 6-8 character is best.Dictionary is exactly for storing The character of verifying password is generated, since the permutation and combination and character quantity of verifying password can be random, the present invention Dictionary in character quantity number it is limited, such as be only number occasion under, dictionary only needs to store 10 numbers just It can be with the character quantity for storing the dictionary of character under normal conditions is corresponding less than or equal to voice messaging of the user in registration The character quantity of content, so that the content for verifying password is as diversified as possible, but the calculating of device with match burden but It is relatively minimal.

(c) to the pretreatments such as voice progress noise reduction, cutting are read aloud, it will specifically read aloud voice and be cut into m sections of voices to be verified Segment, the content of each sound bite to be verified are a character, it is assumed that the number of segment that voice messaging is cut when registering is n Section, then m≤n, n >=1.For example, verifying password content be 8 characters when, read aloud here voice will be cut into 8 sections it is to be tested Demonstrate,prove sound bite；When sound bite to be verified cuts failure, user to be verified is reminded, and resurvey the bright of user to be verified Read voice.

(e) it extracts and reads aloud content and phonetic feature to be verified in voice, according to the model library and verifying established before Password verifies content and phonetic feature to be verified, and is verified result.Particularly calculate user to be verified with The matching degree of data in model library, if being then verified in threshold range；If not in range, then current The verifying of user to be verified does not pass through.

Here two embodiments are specifically related to.

In one embodiment, the present apparatus extracts the phonetic feature of each sound bite to be verified；Then by model The sound bite of user is corresponded in library to be combined according to the content of verifying password, and with the sequence after combining correspondingly with The sound bite to be verified for reading aloud voice carries out phonetic feature matching.When fewer in number, for above-mentioned one four mouthfuls, from It picks out corresponding phonetic feature according to the content of verifying password in the sonification model of four people to be combined, then by everyone The phonetic feature of each content be compared with the phonetic feature of corresponding segment to be verified, matching look at specifically which User, if being matched to corresponding user, successful match, otherwise it fails to match., can be with if if number is numerous in model library By some pre-processings, sequencing of similarity is carried out, picks out corresponding data in the sonification model high from similarity-rough set It is matched, sees whether to be who specific user in model library.

In the second embodiment, we can also first verify user the content for reading aloud voice whether with verifying password Content is identical；If identical go progress Characteristic Contrast then to prompt user to be read aloud again to acquire and read aloud voice if different again, If repeatedly all different, reach preset errors number, then can carry out alarm or other processing.

When specific comparison, it is only necessary to comparison read aloud voice sound bite to be verified it is corresponding be that who is used in model library Family does not need to go the phonetic feature in built-up pattern library in order again.The sound bite and mould to be verified of voice will be read aloud The speech model of correspondence user in type library carries out phonetic feature matching (detailed to be identical with the first embodiment), if being matched to pair The user answered, then successful match, otherwise it fails to match.

When user to be verified is matched to corresponding user, according to the permission that this user is possessed, device can be assigned The corresponding permission of the user to be verified.

In order to realize that the above method, the device that the present invention is previously mentioned may include with lower module: when being registered for acquiring The sound acquisition module (such as microphone) for reading aloud voice when voice messaging and verifying, for extracting voice messaging and reading aloud The characteristic extracting module of the phonetic feature of voice, for the memory module in storage model library and character, the random character that extracts is generated Verify password secret generation module, to read aloud voice and verify password carry out matching verifying authentication module.Wherein memory It is the hardware of device itself, therefore, model library and dictionary are local database, do not need to carry out networking operation.Pass through this The method for recognizing sound-groove of ground processed offline, the content identified every time change at random, the significant increase safety of user's checking.

In addition to specific device, method of the invention is realized based on computer program, and the present invention also protects computer equipment With storage equipment, computer equipment include memory, processor and storage on a memory and the meter that can run on a processor Verification method of the invention may be implemented when executing computer program in calculation machine program, processor.It is equally executable comprising computer The storage equipment of instruction realizes verification method of the invention when executing computer executable instructions by computer processor.

The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.