LG Uplus announced on the 26th that it will incorporate 'Anti-DeepVoice' technology into its artificial intelligence (AI) currency app 'ixi-O'. This is the world's first case of Anti-DeepVoice technology being offered in an on-device format.
ixi-O is an on-device based app that can implement AI on the device without an internet connection. It utilizes 'ixi-GEN,' which was newly developed by training the communication data on the colossal language model (LLM) 'EXAONE' from LG AI Research.
The newly added 'Anti-DeepVoice' in ixi-O is a technology that identifies voices manipulated by AI. A representative from LG Uplus noted, "This technology is gaining significant attention as AI is increasingly used in voice phishing crimes, such as deceiving acquaintances with counterfeit voices or disguising as hostage situations."
In order to develop the Anti-DeepVoice technology, LG Uplus trained its AI engine with approximately 3,000 hours of currency data. This 3,000 hours corresponds to about 2 million currency transactions. It determines the authenticity through processes such as detecting unnatural pronunciations in counterfeit voices or identifying abnormal patterns in voice frequencies.
A representative from LG Uplus stated, "Through ixi-O, it can instantly determine if the voice of the person on the line has been manipulated," adding that "the currency data required to judge whether the voice has been altered is about 5 seconds long."
Anti-DeepVoice is implemented using technologies developed by LG Uplus, including ▲Voice Activity Detection (VAD) ▲Speech-to-Text (STT) ▲Anti-spoofing analysis. It detects actual voice segments during the currency, converts the detected voice to text, and then analyzes whether the voice has been manipulated through anti-spoofing. If the voice of the currency partner is found to be manipulated during this process, it alerts the customer of the danger through a pop-up notification.
LG Uplus has also been working on 'lightweighting' Anti-DeepVoice to commercialize it. This was a process to embed it into ixi-O, which operates on an on-device AI basis.
LG Uplus has developed 'Anti-Deepfake' technology to prevent crimes involving synthetic faces. This technology analyzes videos or images to detect unnatural traces remaining in the synthesized video, determining whether it has been synthesized. It detects synthesis by analyzing ▲Pixel-level texture ▲'Pattern imbalance' that is not visible to the human eye but leaves traces ▲Phenomena where inter-frame consistency decreases. In the future, LG Uplus plans to use Anti-Deepfake technology to block phishing crimes using video and to quickly identify manipulated harmful content, thus ensuring customer safety.
LG Uplus plans to secure various technologies to block voice phishing risks that can occur at all stages before, during, and after a currency through ixi-O. It is preparing to introduce an 'AI voice phishing detection system before currency' that automatically detects phone numbers with a high likelihood of voice phishing after the AI learns the currency patterns of reported numbers. It is also developing a 'criminal voice detection system' in collaboration with the National Forensic Service to detect the voices of voice phishing criminals during currency.
Jeon Byeong-gi, head of the AX Technology Group at LG Uplus, said, "We will establish ourselves as 'the carrier with the safest AI and the most genuine commitment to security' through differentiated security technologies."