Back To Top

S. Korea launches higher-accuracy AI-based voice analysis model to tackle voice scams

(123rf)
(123rf)

The government has launched a higher-accuracy voice analysis model aimed at curbing rampant voice phishing scams and other voice-based crimes and plans to share the technology with foreign investigative bodies, a senior interior ministry official said Thursday.

South Korea sees more than 20,000 cases of voice phishing scams every year, wrecking damage worth about 700 billion won (US$537.6 million). A total of more than 156,000 voice phishing crimes were reported during the 2018-2022 period, with their combined damage reaching 3.06 trillion won, according to police data.

The National Forensic Service had relied on voice analysis models developed by Britain and Russia for data analysis in voice phishing investigation, but the foreign language-based tools' accuracy rate in identifying scammers reached only 30 percent. That made the government take on a project to build an indigenous model based on data involving Korean speakers.

The new voice analysis model, which went into service in late February, uses the deep-learning technology to allow analyzers to identify voice phishing scammers by comparing their voice to the voice data pool involving voice phishing offenders, said Kim Chul, director of the ministry's Integrated Data Analysis Center.

The center collaborated with national forensic researchers for the development of the new model.

Some 1 million voice files from about 6,000 Korean and foreign language speakers were utilized for the indigenous model to study voice analysis.

The analysis model, in particular, features for the first time in the world the capacity to identify voice phishing scammers into scam groups so as to allow investigators to identify criminal voice scam organizations in a speedier manner, Kim said.

The model, moreover, can be used for voice analysis in other crimes using the telephone, such as home rental or coin investment frauds as well as false emergency calls.

Kim said the new model's accuracy in identifying voice phishing scammers reached 51 percent in two previous rounds of verification test while the corresponding rate for the existing models was 28.7 percent.

The interior ministry plans to have police install the new voice analysis model in their investigation system nationwide in the second half of this year, a step that is expected to help police apprehend voice scam suspects in a speedier manner and confirm whether they are responsible for extra crimes.

Currently, the ministry is sharing the technology with Indonesia and Mongolia through official development assistance programs, and is willing to share and spread it to other nations by introducing it at international conferences, Kim said. (Yonhap)

MOST POPULAR
LATEST NEWS
subscribe
지나쌤