MUMBAI, India, June 22 -- Intellectual Property India has published a patent application (202621049786 A) filed by Sanket Sanjay Thakare on April 19, 2026, for A Hybrid Deepfake Audio Detection System Using Self-Supervised Embeddings, Gan-Based Anomaly Detection, And Multi-Expert Fusion.
Inventors include Sanket Sanjay Thakare; Sajal Patil; Raj Tambat; Anubhav Singh; and Rupali Mahajan.
The application for the patent was published on June 12, 2026, under issue no. 24/2026.
Abstract: The present invention relates to a hybrid deepfake audio detection system designed to improve robustness and generalization across diverse datasets. The system comprises a preprocessing module for processing input audio, a self-supervised embedding module for extracting high-level speech representations, a GAN-based anomaly detection module for identifying deviations from genuine speech patterns, and a spectral analysis module for detecting artifacts in audio signals. The outputs of these modules are combined using a dynamic mixture of experts fusion mechanism, which assigns adaptive weights based on input characteristics. The system effectively detects both known and previously unseen deepfake audio samples, thereby enhancing the reliability and security of voice-based systems, including biometric authentication and digital communication platforms.
Disclaimer: Curated by HT Syndication.
