MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202611104897 A) filed by Panipat Institute Of Engineering And Technology Piet on August 31, 2026, for Codec-Aware System And Method For Voice Deepfake Forensic Analysis Using Generative Fingerprinting And Evidence Reliability Scoring.
Inventors include Mr. Anurag Vashist; and Mr. Piyush Babele.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: The present invention provides a codec-aware system and computer-implemented method for forensic analysis of natural, replayed, converted, cloned, modified, or synthetic voice data. Received voice data and metadata are preserved and analysed to determine codec, compression, packet-loss, jitter, transcoding, bandwidth, resampling, and timing conditions represented by a channel-degradation profile. A counterfactual reconstruction engine generates multiple compensated audio representations according to the profile. Phase, temporal, spectral, bispectral, excitation-source, neural-vocoder, statistical, and learned features are extracted and compared across the representations to identify stable residual features forming a channel-invariant generative fingerprint. The fingerprint is used for voice-origin classification and known or unknown generator-family attribution. A dual-score controller independently determines a manipulation probability and a forensic evidence reliability score based on technical evidence-quality parameters. A reliability-aware response may be generated, while original audio, metadata, timestamps, processing information, analysis results, model information, and integrity data are cryptographically preserved in a reproducible forensic evidence package. Accompanied Drawing [FIG. 1]
Disclaimer: Curated by HT Syndication.
