Russia Is Building a Service to Hunt Down Voice Deepfakes

Researchers are developing an AI tool to detect synthetic speech in as little as 20 seconds — a move that could reshape the fight against digital fraud.

Russia is preparing to launch a new service aimed at combating voice-based deepfakes, according to State Duma deputy Anton Gorelkin. The system, currently in development by Russian scientists, uses advanced algorithms to analyze audio streams for signs of AI-generated speech — a tactic increasingly exploited in cybercrime schemes. Fraudsters often clone the voices and faces of corporate executives and government officials to deceive their targets.

The planned tool could prove especially valuable in banking, where call center operators need to instantly distinguish between genuine customers and scammers armed with synthetic audio. To train the neural network, researchers fed it more than 200,000 audio samples. The system examines cepstral coefficients — mathematical markers of an audio signal — over a 20- to 40-second window to determine authenticity.

Once launched, the neural network will be maintained and upgraded by a dedicated lab. The project has been nicknamed “Sipuha,” after a barn owl known for its exceptional hearing. Developers hope the service will become a key asset in countering emerging threats to digital security.