Screengrabs of the video sent to the DAU tipline

The Deepfakes Analysis Unit (DAU) analysed a short video clip in which a man sporting a scarf bearing Bharatiya Janata Party (BJP) insignia can be heard speaking in Tamil. An online search revealed that a similar clip featured recently in The New York Times article about A.I. tools being used in India’s elections. An A.I.-startup quoted in that article confirmed to the DAU that they had used A.I. to create the clip that we were reviewing.

The three-second video was sent to the DAU tipline for verification. We paid close attention to the mouth and the eyes of the man speaking in the video. The movement of his lips and mouth was perfectly in sync with his words and the movement of his eyes seemed natural as well. The audio quality was noise-free and clear.

Our fact-checking partners helped us identify the language being spoken in the video as Tamil. “My name is Shakti Singh. It was a pleasure to meet you,” is the English translation of the words heard in the video. We found nothing odd about the intonation or the pronunciation in that video.

A keyword search guided by the name mentioned in the video and a reverse image search of the frames from the video helped us locate The New York Times story, which mentioned The Indian Deepfaker, an A.I. startup, that had created the A.I. likeness of Shakti Singh Rathore, a BJP member from Rajasthan.

Divyendra Singh Jadoun, founder of The Indian Deepfaker, told the DAU that he had used A.I. tools to create a demo video of the politician speaking in four languages— English, Sanskrit, Bangla, and Tamil.

The politician’s voice in Hindi was recorded for about two minutes, and that was used as a sample to clone his voice in different languages using A.I. tools. Separately a video of Mr. Rathore was shot to capture his natural lip movements and facial data.

The audio and the video were then put through A.I. tools to create seamless videos of him speaking in different languages without any inconsistencies in the lip sync. Mr. Jadoun said that lip sync algorithms are used to create such videos.

(Written by Debraj Sarkar and edited by Pamposh Raina.)

Kindly Note: The manipulated audio/video files that we receive on our tipline are not embedded in our assessment reports because we do not intend to contribute to their virality.