The Deepfakes Analysis Unit (DAU) analysed a video that apparently shows India's Finance Minister Nirmala Sitharaman, and Palki Sharma Upadhyay, a television journalist, promoting a financial investment platform supposedly backed by the government. After running the video through A.I. detection tools and getting our expert partners to weigh in, we were able to conclude that the video was manipulated using A.I.-generated audio.
A Facebook link to the two-minute-and-26-seconds video in English was sent to the DAU tipline for assessment multiple times. The video, embedded in a post, was published on May 20, 2025 from an account with the display name of “Womennote”, and a display picture of the name broken into two using a different typeface. In their profile details they mention “beauty, cosmetic, and personal care”.
We don’t have any evidence to suggest that this suspicious video originated from the aforementioned account or another. The video is not available on the same link anymore.
The video has been packaged like a television news segment. It opens with a split screen featuring a medium close-up of Mrs. Upadhyay and Ms. Sitharaman on the left and right side of the screen, respectively. A still image of the latter appears for about 26 seconds while the former can be seen in a studio-like setting apparently speaking to the camera . After that the video cuts to a single clip featuring Sitharaman, which continues till the end.
Sitharaman’s backdrop and clothing in the still image and the video clip are different. In the clip, she can be seen seated with a microphone visible in the foreground, the backdrop is blurred. She has been filmed in a medium close-up in the clip as well, which seems to be an interview as she is not looking into the camera but focusing on something or someone else.
Bold, static text graphics in English appear, throughout, at the bottom of the video frame with the numbers highlighted in yellow and everything else in white. The text reads: “Over 1,000,000 rupees was paid out to every Indian citizen who invested 21,000 rupees in the government platform”.
A female voice recorded over Upadhyay’s video track conveys a sense of urgency. It claims that while Indians are becoming “wealthy” through the supposed platform, the viewers of the video lost an opportunity to earn a million Indian rupees last month. It further states that the “project” is being launched by Sitharaman as part of some “National Artificial Intelligence Development Program”. The monologue ends by urging the viewers to join the purported project with an investment of 21,000 rupees before the three-day registration window closes.
A different female voice has been recorded with Sitharaman’s video track. It too repeats the numbers mentioned in the previous segment and those highlighted through the text graphics. It states that the purported project is part of some “Digital Progress Initiative” and “is the first official and licensed platform in India”, touting that “it is available for everyone”.
The voice adds that the goal of the supposed initiative is “to eliminate poverty and financial hardship in India”, and that people would be able to generate “passive income”. It also boasts that the lives of “many people” have changed through this supposed project, which “works fully automatically”, and is the “most profitable” and “effective program” in the “market” with “guaranteed results”. Tall, unsubstantiated claims about all “registration records” being “broken” are made by that voice as well.
The voice also uses the purported speaker’s position as a goad for people to invest in the get-rich-quick scheme. It states: “my name is well known across the country, and my reputation is impeccable.” To add credence to the platform, the voice announces that, “this is not a scam”. The sense of urgency manifests through this voice as well as it declares that, “smart people don’t waste time” and cautions that the number of registration slots is limited.
The video track in Upadhyay’s clip does not use any transitions. However, white flashes and jump cuts punctuate the flow of Sitharaman’s visuals, resulting in her head changing positions abruptly at different points in the video. This establishes that her video track has been created by editing out certain portions from a source video and stitching the rest together.
Upadhyay’s lip movements appear somewhat puppet-like but align fairly with the accompanying audio track. Sitharaman’s lips move unnaturally fast and sync imperfectly with the audio and the color of her lips also seems to change in some frames in the video. The dentition of Upadhyay appears well captured in the video. However, Sitharaman’s teeth seem to change color and shape; her upper set is present in some frames and absent in others.
On comparing the voice attributed to Upadhyay and Sitharaman, respectively, with that heard in their recorded videos available online, some resemblance can be established between their real and purported voice. The accent of the voice with Upadhyay’s video track compares well with her real accent, however, it sounds foreign when pronouncing Sitharaman’s name. The voice being associated with Sitharaman has a peculiar accent that does not match her natural accent.
The intonation and pauses characteristic of human speech are missing from both audio tracks, making the overall delivery for both voices sound scripted.
We undertook a reverse image search using screenshots from the video. Sitharaman’s clips were traced to this video published on May 19, 2024 from the official YouTube channel of Think School, an education start-up in India. The still image of hers seen in the first segment of the video has been used on multiple websites, such as here and here; it has been in circulation at least since December 2022, we are not sure of the original source.
Despite a thorough reverse image and keyword search we were unable to trace the original video featuring Upadhyay.
The clothing and body language of Sitharaman in the video we reviewed and the video we located on YouTube are identical. However, the backdrop in the two videos looks slightly different because the clips of Sitharaman used in the manipulated version are more zoomed-in with portions of the background cropped out. Upadhyay is not part of the video we traced.
The interviewer seen in the source video does not feature anywhere in the manipulated version, which has been created by stitching together a few clips that feature only Sitharaman. She speaks in English in the original video too but the audio tracks of the two videos are different. The original carries no text graphics but bears a logo on the lower right side of the video frame.
The packaging, tone, and messaging of this video is similar to the slew of financial scam videos that we have debunked. Many such videos have targeted Sitharaman, and have recommended an investment of “21,000 rupees” in the supposed financial platforms or schemes being peddled through the videos.
To discern the extent of A.I. manipulation in the video under review, we put it through A.I. detection tools.
The voice tool of Hiya, a company that specialises in artificial intelligence solutions for voice safety, indicated that there is a 99 percent probability of the audio track in the video having been generated or modified using A.I.
Hive AI’s deepfake video detection tool highlighted a few markers of A.I. manipulation in the portion featuring Upadhyay but did not point to any elements of A.I. in Sitharaman’s segment. Their audio detection tool, however, indicated that the entire audio track of the video is A.I.-generated.
We also ran the audio track from the video through Deepfake-O-Meter, an open platform developed by Media Forensics Lab (MDFL) at UB for detection of A.I.-generated image, video, and audio. The tool provides a selection of classifiers that can be used to analyse media files.
We chose six audio detectors, out of which two indicated that it was highly likely that the audio track in the video is A.I.-generated. AASIST (2021) and RawNet2 (2021) are designed to detect audio impersonations, voice clones, replay attacks, and other forms of audio spoofs. The Linear Frequency Cepstral Coefficient (LFCC) - Light Convolutional Neural Network (LCNN) 2021 model classifies genuine versus synthetic speech to detect audio deepfakes.
RawNet3 (2023) allows for nuanced detection of synthetic audio while RawNet2-Vocoder (2025) is useful in identifying synthesised speech. Whisper (2023) is designed to analyse synthetic human voices.
For a further analysis on the audio track we also ran it through the A.I. speech classifier of ElevenLabs, a company specialising in voice A.I. research and deployment. The results that returned indicated that it was “very likely” that the audio track in the video was generated using their platform.
We reached out to ElevenLabs for a comment on the analysis. They told us that based on technical signals analysed by them they were able to confirm that the audio track in the video is A.I.-generated. They added that they have taken action against the individuals who misused their tools to hold them accountable.
For expert analysis, we escalated the video to our detection partner ConTrailsAI, a Bangalore-based start-up with its own A.I. tools for detection of audio and video spoofs. The team ran the video through audio and video detection models, and concluded that the audio track is A.I.-generated and the video track is A.I.-manipulated.
In their report, they noted that the lip movements of both Upadhyay and Sitharaman look animated and unnatural, and there are certain words where the lip movements do not sync with the speech. They added that this is definitely a lip-sync attack.
The team explained that the voices in the video sound similar to the actual voices of the featured subjects, however, they are slightly monotonous and the pacing is unnatural in certain parts. They indicated that it is most likely an advanced voice cloning or voice conversion attack.
In the video analysis graphs below, “person 1” refers to Upadhyay and “person 2” refers to Sitharaman. The green portion in the graph for Upadhyay’s video track is pointing to Sitharaman’s static image from the first segment of the video, which has been identified as authentic by Contrails’ video detection model.
To get further expert analysis on the video, we escalated it to the Global Deepfake Detection System (GODDS), a detection system set up by Northwestern University’s Security & AI Lab (NSAIL). The video was analysed by two human analysts, run through 22 deepfake detection algorithms for video analysis, and seven deepfake detection algorithms for audio analysis.
Of the 22 predictive models, six gave a higher probability of the video being fake and the remaining 16 gave a lower probability of the video being fake. Six of the seven predictive models gave a high confidence score to the audio being fake and only one model gave a low confidence score to the audio being fake.
In their report, the team noted that occasionally, Sitharaman’s mouth movements and speech do not fully align, resulting in an unnatural appearance. They observed that the shape and alignment of her mouth and nose shift frequently as she appears to speak. They also stated that the voice with her video track lacks natural variation and cadence, making it sound robotic and unnatural.
The team corroborated our observations about the flashes of a white screen in the video, which likely indicate cuts in the footage and suggest that the video has been edited and isn’t entirely authentic.
They even pointed to a specific moment in the video when Sitharaman’s hand appears to pass through her face, which is a further indicator of manipulation in the video. The GODDS team concluded that the video is likely generated with artificial intelligence.
Based on our findings and analyses from experts, we can conclude that separate video tracks of Upadhyay and Sitharaman have been stitched together and used with synthetic audio to fabricate the video. This is yet another attempt to promote a dubious financial investment platform by falsely linking it to the finance minister to scam the public.
(Written by Debraj Sarkar and Rahul Adhikari, 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.
.jpg)
.jpg)
