In April, political controversy erupted in the southern Indian state of Tamil Nadu. K Annamalai, leader of India’s ruling Bharatiya Janata Party (BJP), has released a controversial audio recording of Dravida Munetra Kazagam MP Paranivel Thiagarajan. DMK) is currently in power in the state.
26 seconds low quality audio tapeThiagarajan, who was Tamil Nadu’s finance minister at the time, allegedly accused the party of illegally amassing $3.6 billion. Thiagarajan vehemently denied the authenticity of the recordings, claiming they were “fabricated” and “machine-generated.”
“Never trust audio clips without a specific source.” Chiagarajan tweeted He claimed on April 22: Now it’s easy to make Quoting a news clip about Drake and The Weeknd’s infamous AI-generated song, he spoke up.
On April 25, Annamalai second clip — 56 seconds long and in much clearer audio — Mr. Thiagarajan allegedly despised his own party and praised the Bharatiya Janata Party. Tiagarajan has now called it a desperate attempt by a “blackmail gang” to create a political rift within his own party, with no one claiming ownership of the clip’s source. said. The BJP’s Hindu nationalist politics are largely unacceptable in the southern states of India, and the party is trying to expand into Tamil Nadu through an aggressive campaign. The alleged audio leak is part of a longer list of what is now known as DMKFiles, a series of alleged corruption scandals in which Annamalai has accused the ruling party. He recently promised to release more such files.
Experts have laid out several worrying scenarios of how AI could affect politics, but in India this is a ‘liar’s dividend’, where those in power are unflattering. It may be the first high-profile case of the ability to claim video as plausible and refutable.Deepfake expert talks rest of the world The rise of AI is being used as a ruse to instill information uncertainty in the new political age. They argue that, on the one hand, generative AI has the potential to tarnish reputations and manipulate public opinion, but on the other hand, the technology offers ways to avoid liability by ignoring incriminating evidence as fake. said it could be.
“The bigger threat now is that everyone has plausible deniability.”
“Generative AI/deepfake/synthetic media content is not only increasing, but also , and even if it’s not, there’s a growing number of claims that it’s fake.” rest of the world by email. Gregory pointed to an example in Myanmar where the military challenged genuine evidence of human rights abuses as fake. Most recently, in May, the Twitter account of Mohamed Hamdan Dagalo, leader of the Sudan Emergency Relief Force, posted an audio recording attributed to the general. Rumors of D’Agallo’s death spread online, prompting social media users to speculate that the voice was generated by an AI. However, forensic analysis conducted by witnesses revealed that there was an Arabic recording. very likely genuine.
rest of the world We have shared two audio clips released by Annamalai with the Deepfake Rapid Response Unit for forensic analysis. An effort by Witness to connect his local network of journalists and fact-checkers with his forensic and deepfake experts in the major media. Deepfakes Rapid Response Force facilitated her three independent tests of the clip. Analysts were divided on whether they decided the quality of the first clip was too low to draw any conclusions, or whether they decided the clip was “very likely fake.” rice field. However, we all agreed and considered the second clip to be authentic.
Rijul Gupta, CEO of AI generation and detection firm Deep Media, said, “It would be extremely difficult for current text-to-speech systems to generate this South Indian accented English voice with this fidelity.” said. rest of the world regarding second clip. “The audio does not contain any artifacts related to speech exchange algorithms (e.g. speech-to-speech instead of text-to-speech). For these reasons, our AI experts conclude that clip 2 is genuine speech. ‘,’ he said in an email to the DeepMedia team. DeepMedia’s deepfake detector is currently being used by the US Department of Defense to verify the authenticity of internet media coming out of Russia and China.
After human evaluation, DeepMedia ran the clips through their own deepfake detection tool and showed an 87% confidence level that the second clip was genuine. However, DeepMedia’s rating is first clip Came back as inconclusive. “The level of noise in this clip makes it difficult to determine its authenticity. The audio sounds as if it was AI-generated, but at this time we cannot conclusively confirm or deny its authenticity. You can’t,” the research team said.
According to a team of six deepfake researchers at the University of Naples Federico II, led by Luisa Verdriva, “the second clip is believed to be real.” [under the threshold] for the entire duration of the clip. They collected 50 minutes of Chiagarajan’s live audio from three of his speeches uploaded to YouTube, including one at the Oxford Union. I then compared the audio to two of his clips. “The first clip turns out to be authentic as the first segment, then the distance increases. said in an email.
“Tools that detect manipulation of synthetic media are not available to those who need it most.”
A third evaluation by deepfake detection company Reality Defender, led by Ali Shahriyari, used in-house speech detection models and human experts. The first clip is “very likely fake”. According to the researchers, “some spoken words are missing, as if the speech converter didn’t pick up the correct pronunciation (or did this on purpose to give the impression of authenticity).” It says. Reality Defender also concluded that the second clip was likely authentic. “The strong naturalness of the speech content (e.g. broken words with filler words that make sense in conversation but don’t make perfect sense when written down), emotional changes that are difficult to model with a fake speech generator, and overall good speech quality content (such as native pronunciation),” the team noted in an email.
rest of the world Tiagarajan’s and Annamalai’s offices were contacted but did not receive a response by the time of publication.
Tiagarajan is his unique voice analysis He cited the audio distortion in the first clip as evidence of a deepfake. he also Story of 2020 by Vice — There, an Indian politician replicated his voice and made an election campaign video in English and Hariyambi. This was proof of how misleading video and audio can be created. In a second statement, Tiagarajan said, Said BJP was “using advanced technology and cheap tactics such as exposing these fabricated voices to sabotage our good work.”
Now, with the explosion of generative AI, the public is second-guessing any evidence and wondering if it’s real or fake. “The bigger threat now is that everyone is making plausible denials, right?” Deep Media’s Gupta said. “Anyone can say it’s fake, it never happened. Even if it were true, their supporters might believe it. That’s insanely problematic.” The “liar’s dividend” mentality puts a lot of pressure on journalists and fact-checkers who now have to choose which stories they want to track and verify, Gregory said.
In May, an AI-edited image of two Indian wrestlers smiling as they were arrested for sexually harassing a Bharatiya Janata politician went viral online. Social media users reportedly claimed the wrestlers weren’t serious about the protest.
“There is a detection fairness gap in the world,” Gregory said. “Tools to detect manipulation of synthetic media are not available to the people who need them most. They have not evolved with the skills to use them into broader journalists and fact-checkers on a global scale.” , Gupta expects more independent companies to develop detection tools as generative AI evolves. “It’s going to be a cat-and-mouse game.”
