Li Fan is the head of the voice recording technology of the Feiyun Platform Business Unit of Keda News. As the person in charge of R&D, he participated in the construction of the “Xunfei Open Platformâ€, which provides the voice interactive service capability for the billion-dollar Internet users, and is responsible for the platform authentication service line technology. Incubation and ability output. The following is the content that Li Fan recently shared in an open class:
Good evening everyone, I am Li Fan of the University of Science and Technology. Keda Xunfei has been conducting basic research in voice technology for more than 16 years. Currently, it is building an artificial intelligence ecosystem. It has an artificial intelligence open platform that is open to the Internet and outputs cutting-edge technologies.
At present, biometrics technology has been gradually promoted and used in the financial field. In the previous issue, Titans shared the application of face recognition technology (titanium dry goods) and iris recognition (titanium dry goods) in the financial field, and also explained why these biological characteristics can be applied in the financial field. Today, I mainly explain from the perspective of technical solutions. A multi-biometric fusion authentication system is an authentication solution based on face recognition technology and voiceprint recognition technology, which can simultaneously perform fusion authentication for multiple biometrics.
Earlier, the International Biometrics Organization conducted a systematic statistical analysis of the biometric features in use in the market from the four levels of interference user level, independence, cost and ease of use, including palmprint recognition, signature recognition, Fingerprint recognition, speech recognition, iris recognition, retina recognition, face recognition, and thermo-spectral recognition. It is proposed that an ideal biometric system needs to have high uniqueness, high ease of use, low degree of user interference and low cost. It can be seen from this that the voiceprint recognition, face recognition technology and signature recognition technology within the scope of human cognition, as well as fingerprint recognition outside the scope of human cognition, iris recognition, and palmprint recognition, can be in their respective fields. Play an important role.
Voiceprint recognition technology and face recognition technology can be used as a convenient form of identity authentication, which is inseparable from its own characteristics. First of all, these biological features are inherently unique and difficult to be copied. Secondly, these features are easy to collect and the collection is more concealed, and can be remotely operated without adding new hardware costs.
In terms of voiceprint technology, voiceprint is based on the speech parameters that reflect the psychology and behavioral characteristics of the speaker, and automatically recognizes the identity of the speaker. Everyone has a unique voiceprint no matter how similar they are. This is formed by the growth of each person's vocal organs, which uniquely identifies a user's identity.
Xunfei is the earliest manufacturer to start researching voiceprint recognition technology in China and has reached the practical level. It has provided relevant technical support for public security departments and financial institutions before proposing multi-biometric integration certification solutions. As a way of authentication, you can identify your identity in natural speech and have a high degree of acceptance.
The voiceprint recognition mode currently supports voiceprint freedom, dynamic digital passwords, open text passwords, and fixed text passwords. As shown in the figure below, each recognition mode has its corresponding application scenario. Dynamic digital passwords have become the most widely used voiceprint recognition mode in financial payment because of their ease of use.
In this field, the main uses of voiceprint recognition are divided into two types, similar to face recognition. One is voiceprint confirmation, that is, 1:1 alignment, and the second is voiceprint discrimination, that is, 1:N alignment. The voiceprint confirmation directly compares the voiceprint model data of the person to be verified with the voiceprint model data at the time of registration, and obtains a similarity score, and the voiceprint discrimination finds the similarity with the person to be verified in a group of models. The highest voice reporter, of course, whether this similarity score can be used as a basis for judgment, and the application needs to be set according to its tolerance.
Let's talk about the measurement of the effect of voiceprint recognition technology (the essence of this effect measurement is the trade-off between security and ease of use):
False Accept Rate, which refers to the ratio of the number of times that the person is mistakenly considered to be the number of times the voice is taken;
False Reject Rate, which is the ratio of the number of times the person is mistakenly considered to be non-self, and the number of voiceprint confirmation operations;
Equal Error Rate, which is the percentage when FA is equal to FR. It is an important basis for measuring the effect of voiceprint verification.
At this point, everyone must have a basic understanding of voiceprint recognition technology and face recognition technology.
Some people may ask, the current face recognition as a payment authentication method has been promoted in the market, and the general feedback effect is very good, why should we do this multi-biometric fusion? In terms of face recognition technology, under the latest deepID2 algorithm, the resolution of face can reach 99.47%, but in practical applications, it faces various challenges, such as a larger one that everyone can feel. The challenge is, these Korean ladies next to us; of course, the voiceprint recognition technology also has application challenges, such as the sound pattern is easily deformed, susceptible to physical conditions, age, mood, etc., and the actual use will be subject to recording equipment. Impact. Therefore, as a single biometric authentication, there are certain instability and limitations.
In order to cope with the high security requirements in the payment field and solve the instability and limitations of single biometrics, a multi-biometric fusion authentication solution is proposed. The core foundation of multi-biometric fusion is the independence and complementarity between features. The fusion of the two biometrics in different dimensions can maximize the advantages of the two biometrics, and at the same time reduce the misrecognition rate to a certain extent, such as the influence of voiceprint deformation or recording equipment.
The multi-biometric fusion authentication solution is a complete Internet solution with key features including:
1. Unified access mode, cloud service continuous optimization and upgrade, flexible access for new and open biometrics, and integrated applications can flexibly select new cut biometrics according to needs.
2. Customize the security threshold and set the security level in combination with the application scenario. In the field where the security requirements such as payment authentication are high, it can be set that all biometrics must be matched when all biometrics must be matched, and the security requirements are not high. In the field, such as the punching system, it can be set to pass when the matching degree of any one of the biometrics meets the standard. In this way, the pass rate can be effectively improved while ensuring the accuracy of the check.
3. The closed-loop security check can provide closed-loop security check interaction for the application, avoiding tampering or forgery in the feature interaction process, and the feature transmission channel adopts multi-layer encryption.
The application of multi-biometric fusion authentication solution in the field of payment, here are a few examples.
1. In August 2015, HKUST News Alliance and China UnionPay and Huishang Bank released the “Voice + Face†integration certification personal transfer application, which is the first product in the field of mobile payment with multi-biometric fusion technology. Effectively protect against the risk of fraud in picture or video playback. In the process of binding the bank card, the function of automatically identifying the ID card and the bank card is also applied, which significantly simplifies the user's operation.
2. In February of 2016, Huishang Bank officially released the mobile banking version with the face voiceprint recognition function. When using mobile banking, users can use “face + recognition†instead of manually entering the password.
3. A number of securities companies that are currently docking are hoping that this solution can be used for remote account opening and secure payment. Users do not have to go through complicated manual verification audits, nor do they have to worry about missing an investment opportunity due to the complexity of the audit process. Both users and businesses are a more efficient application.
In addition to the financial payment field, this multi-biometric feature authentication solution is also used in other fields. For example, the identity authentication attendance currently used by HKUST is flying, and the branch offices across the country only need to perform the app through the app at the designated location. "Face + voiceprint" can be punched.
Guangzhou Yunge Tianhong Electronic Technology Co., Ltd , http://www.e-cigaretteyfactory.com