Because Desi languages (Hindi, Urdu, Tamil, Bengali) have distinct rhythmic and syllabic structures, generic detectors often fail. VideoDesiFakesNet uses a specialized LSTM (Long Short-Term Memory) model to ensure that the lip movements precisely match the phonemes of the spoken language. A lag of even 20 milliseconds is flagged as suspicious.
Because Desi languages (Hindi, Urdu, Tamil, Bengali) have distinct rhythmic and syllabic structures, generic detectors often fail. VideoDesiFakesNet uses a specialized LSTM (Long Short-Term Memory) model to ensure that the lip movements precisely match the phonemes of the spoken language. A lag of even 20 milliseconds is flagged as suspicious.