The tools, developed by UiB doctoral student Sohail Ahmed Khan, are used for recognizing armored vehicles and for language detection in videos
The tools were developed by Khan during a two-month stay with NORDIS partner Faktisk.no’s verification newsroom, Faktisk Verifiserbar. The newsroom, while headed by Faktisk.no, was a joint effort between a wide range of media companies, using open source intelligence (OSINT) techniques to verify audiovisual material from the war in Ukraine.
The prototypes Khan developed were a «Tank Classifier» and a «Language Checker». The Tank Classifier relies on deep learning to predict the type of a military vehicle present in an image, while the Language Checker uses OpenAI’s Whisper model to transcribe speech in videos, identify the language(s) spoken, and then translating the text to English.
In an interview with Journalisten.no, Khan explains that identifying the language spoken in a video is intended to support journalists attempting to pinpoint the location where a vide was recorded. For this reason, the Language Checker tool also supplies the region in which a detected language is spoken.
The prototypes can be accessed here. Note that predictions are not perfect, and that some errors should be expected.