Home > National > Social Affairs

Prosecutors to develop AI technology to identify fake voices

Published: 25 Mar. 2024, 15:44

KIM JI-YE
[email protected]

The AI voice of late folk singer Kim Kwang-seok (1964-1996) singing Bibi's latest hit ″Bam Yang Gang″ in a video uploaded in February. [SCREEN CAPTURE]

Prosecutors are beginning to deploy artificial intelligence (AI) technology to identify fake voices.

Videos with deepfake voices are becoming popular on social media platforms such as YouTube. In particular, videos that mimic singer's voices by using AI technology are becoming increasingly common.

These videos train AI to sing songs based on a singer's voice data and can create impossible situations, such as having late folk artist Kim Kwang-seok (1964-1996) sing currently active artist Bibi's latest hit "Bam Yang Gang."

Considering that the technology mimicking other people's voices is now accessible for the average person, prosecutors began conducting extensive research anticipating crimes using fake voices.

“We plan to develop an AI program that can detect and classify fake voices over the next four years, starting with this year,” the voice analysis team under the Supreme Prosecutors’ Office’s Forensic Science Investigation Department said last Friday.

The research is a part of the seven-part project the prosecutors’ office aims to complete this year to strengthen its forensic capability.

The project is the first time prosecutors are directly addressing developing technology about fake voices.

Voice data is already in use during investigative and trial processes.

Prosecutors have a voice analysis team that reviews voice data to detect vocalizations and pronunciations. Based on the analysis, the team can distinguish if voices are the same or different, as well as if someone tampered with the data. They also enhance data quality by removing unnecessary noises.

Voice analysis played a significant role in solving a murder and kidnapping case from March last year, when three culprits kidnapped a woman in the Gangnam District, southern Seoul, and later murdered her to steal her cryptocurrency.

The voice analysis team cleaned the voice data from the black box of the car the criminals used to kidnap the victim. The original recording had sounds from the car that muddled the conversation.

Once the conversation became clear, it became crucial evidence proving that the crime was plotted beforehand and overturned the criminals' initial claim that the offense was an accident.

Voice analysis is also used to verify that voice data hasn't been manipulated in order to qualify it as valid evidence. This is especially common to analyze voice phishing cases and recordings of people receiving cash or goods for an election campaign, which violates the Political Funds Act and the Public Official Election Act.

The Supreme Prosecutors’ Office’s voice analysis team will conduct their medium- and long-term projects developing fake voice detection technology from 2024 to 2027.

This year, the voice analysis team will work with companies that have a voice database, such as Naver and KT, to examine voice generating technology and collect information on detecting tampered voice data.

The team already secured government funding for the research and is scheduled to release a bidding announcement in April to select businesses for collaboration.

“Due to the rapid development of AI technology, now anyone can make a fake voice,” Kim Kyung-wha, the head of the voice analysis team, said. “We have been preparing to develop detective technology since last year in order to preemptively act against crimes involving fake voices.”

“We expect that the developed technology will be effective, especially on voice phishing cases that use a family member’s voice.”

BY YANG SU-MIN, KIM JI-YE [[email protected]]