Cheap audio, pic, video cloning tools have made deepfake creation extremely powerful. Yella's cloning detectors flag fake content with just 3-5 secs of audio. Or even one pic (eg a single frame from a video). Our ML models pack a punch in a footprint < 1Gb.
Hello PH Family🎉
Tools to create deepfake audio, pics & video have grown very sophisticated. Yella has built some tools to catch up with these cloning tools:
- Sloka is a cloned audio detector. It works at close to 100% accuracy with a small (< 1Gb) memory footprint. Popular tools like ElevenLabs, Lovo, Descript, Podcastle, Play.ht... and many more are flagged at close to 100% accuracy
- Chitra uses the same underlying principles of Sloka to flag cloned pics. We focus on fake profiles/portraits. Nowadays, sophisticated scammers use a mix of clong tools like MidJourney, Image Creator, DALL-E... And throw in some PhotoSHop-type editing. Chitra flags such doctored headshots/profile pics.
Requirements:
-Just 3-5 seconds of audio
-Just one pic (one frame from a doctored video suffices). Preferably 256x256 px ie normal sized ones
Potential bizs affected by such deepfakes:
Banks, Fintech (Square, PayPal, Fidelity, Charles Schwab...), eCommerce vendors, HealthCare
Hopefully, Yella helps businesses/cinsumers fight deepfake scams🤐