Yella delivers deepfake audio & picture detection tools to secure businesses against modern communication frauds. The ML models in these detectors are based on Dec '23 research with custom IP.
Current Status
MVPs built Feb/Mar 2024. Not yet marketed to businesses.
Problem or Opportunity
In the last year, the FTC has identified cloned audio as the #1 source of consumer & SMB frauds. Cloned images & video have also reached the tipping point where they are indistinguishable from real media. Banks, fintech companies, healthcare, eCommerce vendors, payment vendors, social media & government agencies (SSA, INS, DHS, IRS) are especially vulnerable. Traditional cybersecurity companies like McAfee, Crowdstrike, ZScaler... & KYC/AML ones (Jumio, Onfido..) do not have any products in these areas.
Solution (product or service)
Sloka is a custom ML model built on research that showed 98% accuracy on a small set of commercial & open-source tools. We extended it to handle the top-10 heavily used tools in creating deepfakes (incl. ElevenLabs, Play.ht, Descript, TTSMaker, Narakeet, Lovo.ai). Chitra is an allied cloned image detector built on the same principles of Sloka. It also uses novel ML models to circumvent an observation OpenAI stated - "deep learning image classifiers do not scale". Both Sloka & Chitra are delivered currently as web services. The models are tiny (
Business model
Our focus is b2b. And the sectors we mentioned above. KYC/AML along with voice biometrics users are our primary targets. For smaller vendors, we have a metered subscription model. For larger customers, we have an on-premise hosted model for an annual fee.
