.jpg)
The Deepfakes Analysis Unit (DAU) analysed a video that apparently shows Col. Sofiya Qureshi of the Indian Army making a statement about her faith, allegiance, and intent to combat terrorism. After putting the video through A.I. detection tools and getting our expert partners to weigh in, we were able to conclude that synthetic audio was used to fabricate the video.
The 18-seconds-long video in Hindi was escalated to the DAU, multiple times, by its fact-checking partners for analysis. The video purportedly shows Colonel Qureshi in uniform, making a speech apparently at a public forum as two microphones and a laptop can be seen placed in front of her. A close-up of her face seems to have been filmed from an angle and a brown wall makes up the backdrop.
Bold, static text in Hindi, visible in the lower half of the video frame, throughout, translates to: I am Muslim, but not a Pakistani terrorist. Those words end with a “thumbs up” emoji. Right below, a white rectangular box with her name in bold and a social media handle — “@dinesh_chauhan” — appear, superimposed on the laptop lid.
A female voice recorded over the colonel’s video track invokes Muslim faith, refutes allegiance to Pakistan and terrorism, and states that terrorists have no religion. It boasts of the courage to kill every terrorist with bare hands without bothering about their religious identity. The video ends abruptly, making the colonel’s face appear contorted. There is some background music that can also be heard in the video but we could not identify the particular composition.
The video does not appear to have any cuts or transitions. However, an odd translucent mark resembling the “pause” symbol is visible in the area between her nose and chin in the first frame of the video; it vanishes as soon as the video starts playing.
In some frames square-like blocks appear on her chin and in some other frames pixellation is visible on her lips when they get close to the microphone. Her lip movements are imperfectly synced with the audio track. Her teeth look blurred, lacking natural shape.
The voice attributed to her and that heard in her recorded interviews online does not sound similar. The accent in both the voices is not identical. The mention of the word “jigra”, a Hindi word for courage, is accompanied by a thumping sound, resembling the slapping of a hard surface. Such emotionally charged delivery stands in contrast to the calm and stoic demeanour that comes through in her natural delivery.
We undertook a reverse image search using screenshots from the video and traced the colonel’s clip to this video published on May 7, 2025 from the official YouTube channel of Hindustan Times, an Indian English-language newspaper and news website.
Her uniform is identical in the video we reviewed and the one we were able to trace. However, the background and foreground look slightly different in the two videos as the frames used in the manipulated version are more zoomed-in with portions of the background and foreground cropped out.
She speaks in Hindi in the original video as well. But, the audio track in the two videos is completely different. There is no thumping noise in the original audio nor does it have the background music.
The source video features other subjects as well but none of them are part of the doctored video, which seems to have been created by lifting a clip featuring the colonel. A logo of Hindustan Times can be seen in the upper right corner of the video frame in the original video.
A small text graphic in the top left corner of the original video reads “YT: MEA INDIA” with the date of the broadcast below it. The text seen in the manipulated video does not appear anywhere in the original video nor does the white rectangle with the social media handle, as a result the logo on the laptop in front of the colonel is clearly visible.
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 20 percent probability of the audio track in the video having been generated or modified using A.I.
Hive AI’s deepfake video detection tool did not highlight any markers of A.I. manipulation in the video track. Their audio detection tool also did not detect any A.I. manipulation in the entire audio track of the video.
We would like to mention that if there is background music or noise in an audio track it can impact the results from the tools.
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 seven audio detectors out of which six gave strong indicators of the use of A.I. in the audio. 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) and RawNet2-Vocoder-V1 (2023) are useful in identifying synthesised speech. Whisper (2023) is designed to analyse synthetic human voices.
For a further analysis on the audio track we put 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 unlikely” that the audio track in the video was generated using their platform.
For expert analysis, we escalated the video to our detection partner ConTrailsAI, a Bangalore-based startup 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 there was A.I. manipulation in the video and audio track.
In their report, the team noted that the colonel’s audio clone has been used to synthesise a similar sounding audio. They added that A.I. signals were clearly audible and caught by the detectors as well. They further stated that A.I. has been used to lip-sync the audio with the original video to make it appear more realistic.
To get another expert to weigh in 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 and run through seven deepfake detection algorithms. Of those, four predictive models gave a higher probability of the video being fake and the remaining three indicated a lower probability of the video being fake.
In their report, the team highlighted several signs of A.I. manipulation such as music playing in the background, which could have been added to the video as a strategy to hide media manipulations. They further stated that the subject's lips and teeth merge in one frame, which is a possible indicator of manipulation in the video.
They pointed to a faint “pause button” at the start of the video, which we pointed to as well in our observations above. According to them, it possibly indicates that this is only a small selection of a longer video that may have been created or that the video source is separate from the audio. They added that this could be another sign of inauthenticity.
The GOODS team noted that the social media handle “@Dinesh_Chauhan” tagged in the video is the author of the post which carried the video. The bio of this handle on X describes “satire/sarcasm” as part of their details, the team believes that too could suggest that the video is possibly inauthentic.
The GODDS team concluded that the video is likely generated with artificial intelligence.
To get another expert to review the video featuring the colonel, especially the audio track with it, we reached out to our partner Validia, a San-Francisco based deepfake cybersecurity service. They used their proprietary software to check the authenticity of the video and audio being associated with her.
The team at Validia noted that the video is a very well crafted lip-sync deepfake, indicative primarily by the start of the video track.
To analyse the audio, they compared the audio track from the video with a sample of her real voice and arrived at a similarity score of 0.86. A similarity score of 1 would mean perfect similarity and anything between 0.8 to 0.99 would be a likely deepfake voice.
The team noted that some of their metrics, including a heat-map analysis, did show significant areas of differentiation. However, they believed that it could be due to the background music that was added to the speech audio. Based on all the similarities identified by them, they concluded that the audio is a deepfake.
Based on our findings and analyses from experts, we can conclude that original footage featuring the colonel was used with a deepfake audio to fabricate the video during the India-Pakistan military escalations.
(Written by Debraj Sarkar and Rahul Adhikari, edited by Pamposh Raina.)
Kindly Note: The manipulated video/audio files that we receive on our tipline are not embedded in our assessment reports because we do not intend to contribute to their virality.
You can read below the fact-checks related to this piece published by our partners:
.jpg)
