AI Wave Promotes The Landing Of More Than 100 Billion Voiceprint Market: The Advent Of Personal ID Era

After fingerprint, face and iris, voiceprint, as a part of our personal ID, has also opened a commercial trend.

Recently, industrial and Commercial Bank of China has started to apply voiceprint recognition technology to credit card anti fraud scenarios. It has officially launched its four branches in Beijing, Hubei, Sichuan and Shanxi, taking the lead in opening a new mode of risk management and control in the same industry.

From the simple technology itself, voiceprint recognition has been studied and explored in the 20th century. In recent years, some mature schemes have been adopted by some domestic and foreign banking systems, which is the initial stage of the application of voiceprint recognition technology.

The core to mature is still relying on the breakthrough of AI technology in the 21st century. Especially in the early months of the epidemic background, the market demand for this kind of non-contact and high accuracy technology application suddenly enlarged or even increased day by day.

During the period of sudden increase, there is a great demand for voiceprint recognition technology. Made by Xu Hui

According to the analysis of prospective industry research institute, the global biometric industry is huge at present. The market scale of voice biometric (i.e. voiceprint recognition) is expected to exceed 140 billion yuan (about 20 billion US dollars) in 2020, accounting for 22.4% of the whole biometric market.

With more and more mature voice technology such as 5g and CTO, more and more reporters can report to the 21st century. Voiceprint recognition technology is expected to play an increasingly important role in finance, public security, intelligent hardware, IOT and other fields.

Layer by layer technology evolution

When it comes to voiceprint, it's easy to understand it as a sound line that can be easily imitated. In fact, the two are not equal.

Some industry insiders have told reporters that it is difficult to camouflage voiceprint, because its recognition content mainly includes vocal cord vibration frequency, vocal tract and other features of the voice model, which can be camouflaged only in the catchphrase and other behavioral characteristics.

"The tone, tone and other habits of each person are almost unique. The information is measured and analyzed by an electroacoustic instrument, which constitutes a different spectrum of sound waves. Therefore, it is unique and stable to use voiceprint to authenticate one's identity." He continued.

According to a white paper on voiceprint recognition technology released by the auditory intelligence research center of the Institute of artificial intelligence of Tsinghua University, the mainstream biometric recognition schemes on the market include palmprint, iris, retina, face and voiceprint. The iris and retina are extremely difficult, the voiceprint is difficult to assess, and the rest are general.

Based on these background, the research on voiceprint has already existed, but with the iteration of the underlying technology, the application process can be more insensitive and achieve higher accuracy.

Chen Dongpeng told the 21st century economic reporter that in order to save time and improve the user experience, the early voiceprint recognition technology would require users to read a fixed text, and then each verification would read the same text.

This is called fixed text authentication. "But it will cause two problems. The first is that the user experience is not good, and he is forced to read a text every time, which makes it impossible to complete the identity authentication unconsciously with users in natural communication; the second is that it is easy to be attacked by recording, because every time the content is read, criminals may attack the system by recording other people's voice in advance and then playing back the recording. ”He further said that this would lead to greater risks in the initial technology landing.

Since 2000, voiceprint recognition technology has entered a relatively mature period. During this period, a number of old voice manufacturers were born, among which iFLYTEK is one.

The aforementioned industry personage pointed out to the reporter that the key point of voiceprint recognition technology at this stage is to compress the input sound into a fixed vector for matching, and the accuracy rate has been able to reach about 95%. The disadvantage is that it is easy to affect the results due to external interference, so it has not yet reached the standard suitable for commercial use.

Since 2015, with the deepening of AI technology, many start-up companies with the background of AI technology have devoted themselves to the field of voiceprint recognition. Moreover, in the past two years, bat and other giants have also entered this field.

"Today's progress in voiceprint recognition technology is due to the maturity of deep learning technology, as well as advances in sensors, microphones, and communication channels (such as 5g, 4G volte high-definition calls) Chen Dongpeng said that the current voiceprint recognition technology of Shengyang technology can ensure the robustness of voiceprint recognition in high noise environment, complete speech authentication in a short time, and do not need to set fixed communication content.

Application scenario expansion

Although it can be put into commercial use under the condition of better experience, voiceprint recognition technology still has a long way to go.

Of course, there is still a lot of room for the mature application of voice print in the industry from the current mature stage to the mature stage of the industry.

This is true for many biometric technologies. Even though the module has been widely used for many years, such as fingerprint recognition, it can continue to be widely used in mobile phone identification.

Chen Dongpeng told reporters that the next challenge of voiceprint recognition technology mainly includes two aspects: far-field identification and optimization of cross-channel recognition.

"Although Shengyang technology has been able to recognize people in a far-field of about two meters, our goal is to be like a human ear, which can identify people from a long distance by voice." He further explained.

The so-called acquisition of signals in the early days of the computer, or even more, can be expanded into the future through the microphone. "It will still be affected by the channel. With the progress of communication technology, data accumulation and higher robustness of training model, I believe this problem can be better solved," Chen added

It is also related to the positive embrace of the industry and the introduction of relevant standards.

In 2014, the Ministry of public security formulated standards for the application technical requirements and security level of voiceprint recognition for the first time, and determined the detailed contents of guidelines on voiceprint database construction and information collection in 2019.

In the financial industry, in 2018, the people's Bank of China issued relevant industry standards for voiceprint recognition, and in 2019, the voiceprint recognition system has become the only biometric technology selected in the certification catalog of financial technology products, laying the foundation for today's exploration and implementation.

The application of the anti fraud scenario of bank credit card mentioned above is mainly through telephone banking. When processing credit card application, card opening and due diligence online, personal voiceprint ID files can be set up remotely online and provide basis for subsequent identification and fraud risk judgment.

It can be imagined that during the epidemic period, the demand for non-contact increased suddenly, and voiceprint recognition technology also ushered in more opportunities to land. Chen Dongpeng introduced, such as access control, elevator, return visit platform, etc., can use voice print technology. "We made several related products during the epidemic, such as voiceprint access control system and remote health return visit system. This system was also first used by Nanshan Yuehai Street office."

"One of the advantages of voiceprint is that it is contained in the voice that people need to communicate. These voice sounds contain extremely rich personal information, such as gender, age, region, environment, emotion, occupation, preference, etc. the identity information revealed by voiceprint is the central point to gather these personalized information, which will be the" user-centered "service provided by many industries It must be mastered. " He concluded.