Government Develops World's Leading 'Voice Phishing Voice Analysis Model'

Ministry of the Interior and Safety and National Forensic Service Collaborate... Utilization from the End of This Month
Improvement in Individual Accuracy and Grouping Capability
77% Higher Accuracy Compared to Foreign Products... Potential for Export

[Asia Economy Reporter Kiho Sung] Our government has developed the world’s first ‘Voice Phishing Voice Analysis Model’ focused on detecting phone scams. It features the ability to identify not only whether a single criminal’s voice matches but also to analyze by groups. Additionally, it is the most accurate model globally, raising expectations for export success.

The Ministry of the Interior and Safety announced on the 22nd that it has developed a ‘Voice Phishing Voice Analysis Model’ that can be used to catch voice phishing scammers and will utilize it in investigations such as voice authentication starting at the end of this month.

The Ministry of the Interior and Safety held a policy briefing on the 21st regarding the 'Voice Phishing Voice Analysis Model' that can be used to catch voice phishing scammers. From the left in the photo are Yoon In-sik, Deputy Director of the Integrated Data Analysis Center at the Ministry of the Interior and Safety; Kim Cheol, Director of the Integrated Data Analysis Center at the Ministry of the Interior and Safety; Na Gi-hyun, Head of the Digital Division at the National Forensic Service; and Jeon Ok-yeop, Researcher at the National Forensic Service. Photo by XXX

Until now, in Korea, the National Forensic Service has conducted voice authentication necessary for voice phishing investigations using voice analysis models developed in Russia and the United Kingdom. However, due to the nature of these models being trained in foreign languages, there were limitations in accurately determining whether Korean-speaking criminals were the same individuals.

In particular, because voice phishing criminal organizations operate in groups divided by roles (such as investigators, prosecutors, etc.), a function to cluster criminals involved in crimes was essential, but existing models lacked this capability.

Accordingly, the Ministry of the Interior and Safety’s Integrated Data Analysis Center, together with the National Forensic Service (hereafter NFS), has been working since last year to develop a model focused on catching voice phishing criminals that improves speaker differentiation accuracy and enables grouping of crime-involved individuals.

Utilizing the latest artificial intelligence learning (AI deep learning) technology, this model was developed using over one million foreign and Korean voice data samples extracted from about 6,000 people domestically and internationally. Specifically for Korean, more than 100,000 general public voice data samples and actual voice phishing scammer voice data held by the NFS were used together, repeatedly conducting various training and performance verification processes to create an optimal algorithm necessary for voice phishing speaker differentiation.

Performance verification results confirmed that the model’s accuracy in correctly identifying criminals’ voices improved by approximately 77% compared to existing foreign analysis models. Along with this, the model also achieved the world’s first implementation of a criminal participant grouping function, which was not expected from previous models.

According to recent data released by the National Police Agency, over the past five years, there have been a total of 156,249 voice phishing damage cases in Korea, with losses exceeding 3 trillion won, seriously impacting citizens’ lives.

The Ministry of the Interior and Safety’s Integrated Data Analysis Center plans to actively utilize the model in collaboration with the NFS and the National Police Agency during voice phishing investigations and arrests, and also promote its overseas expansion.

First, the NFS will begin using the newly developed model for voice authentication of voice phishing scammers from the end of February. It will analyze about 10,000 voice phishing scammer voice data samples in its possession to group criminal organizations and investigate additional crimes of already arrested criminals. Furthermore, the model will be shared with the National Police Agency to enhance the speed of initial investigations and arrest rates for voice phishing crimes, and will be applied broadly to various voice-related crimes such as impersonation of institutions and jeonse fraud.

The voice data of voice phishing scammers analyzed using the new model developed by the Ministry of the Interior and Safety’s Integrated Data Analysis Center will also be published on the Financial Supervisory Service’s website (Voice Phishing Guardian) to raise awareness and improve prevention levels against voice phishing crimes.

Meanwhile, the Ministry of the Interior and Safety plans to promote overseas expansion as a new administrative Korean Wave product. For countries wishing to learn our latest voice forensic techniques, including developing countries, the ministry will showcase the model’s excellence and scalability through training and various international events scheduled for the second half of this year.

Han Chang-seop, Vice Minister of the Ministry of the Interior and Safety, evaluated, “The newly developed voice phishing voice analysis model is a concrete outcome of the Digital Platform Government that solves social issues through data analysis,” and added, “Going forward, we will continue to discover analysis tasks needed by the public and actively apply analyzed results in the field to realize a data-driven, efficient government.”