AI Engine Trained on 3,000 Hours of Call Data
Technology Secured to Detect Even Deepfake Videos
On June 26, LG Uplus announced that it will commercialize a service that uses artificial intelligence (AI) technology to detect manipulated voices in real time.
On this day, LG Uplus revealed that it will integrate the 'Anti-DeepVoice' technology?first unveiled at Mobile World Congress (MWC) in March?into its AI call agent service 'ixi-O', launching the service on June 30. According to LG Uplus, this marks the world's first commercialization of such technology in an 'on-device' format, where customer information is stored only on the customer's smartphone, not on the telecom server.
Anti-DeepVoice is a technology that filters out AI-manipulated voices, developed in response to the recent increase in AI-based voice phishing crimes. To develop this technology, LG Uplus trained its AI engine with approximately 3,000 hours of call data, amounting to about 2 million calls.
The completed Anti-DeepVoice technology detects unnatural pronunciations or abnormal patterns in voice frequencies that are characteristic of forged voices. LG Uplus explained that only about five seconds of call data are needed to determine whether a voice has been manipulated. If the system determines that the caller's voice has been altered, ixi-O will display a popup notification.
Following its voice phishing detection capabilities, LG Uplus has also secured 'Anti-Deepfake' technology, which analyzes AI-synthesized faces. This technology analyzes videos or images to detect unnatural traces left in synthesized footage and determines whether the content has been manipulated.
LG Uplus is also developing additional technologies through ixi-O to block voice phishing risks that may arise before, during, and after calls. The 'Pre-call AI Voice Phishing Detection System' automatically identifies risky numbers before a call begins by having AI learn the call patterns of phone numbers reported for voice phishing. Additionally, the 'Criminal Voice Detection System', currently being developed in collaboration with the National Forensic Service, compares and analyzes the voiceprints of voice phishing criminals held by the National Forensic Service in real time using AI. If there is a high likelihood of a match, the system issues a warning notification.
Jeon Byungki, head of LG Uplus AX Technology Group, stated, "With differentiated security technology, we aim to establish ourselves as the telecom company with the safest AI and the most genuine commitment to security."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
