MUMBAI, India, Aug. 17 -- Intellectual Property India has published a patent application (202621074071 A) filed by Ashokrao Mane Group Of Institutions, An Autonomous Institute, Kolhapur on June 15, 2026, for “ A System And Method For Deep Fake Audio Detection Using Mel Frequency Cepstral Coefficients And Ensemble Deep Learning Architectures ”.

Inventors include Prof. Pravin Kumar Karve; Prof. Prathamesh S. Powar; Prof. Vikas A. Patil; Prof. Suhas S. Kibile; Prof. Almas A. Mahaldar; and Ms. Payal S. Patil.

The application for the patent was published on August 14, 2026, under issue no. 33/2026.

Abstract: The invention discloses a system and method for detecting deep fake or synthetically generated audio content using a multi-feature extraction framework integrated with ensemble deep learning architectures. The system receives input audio signals and preprocesses them through noise reduction, amplitude normalization, resampling, and frame segmentation operations. A multi-feature extraction module simultaneously computes Mel Frequency Cepstral Coefficients (40 coefficients with delta and delta-delta derivatives), mel-spectrogram representations (128 mel bands), and temporal acoustic features (zero crossing rate, RMS energy, spectral centroid, spectral bandwidth, and pitch contour) to capture comprehensive spectral, temporal, and perceptual characteristics. An ensemble classification module processes these features through parallel CNN, bidirectional LSTM, and ResNet classifiers, fusing their outputs through a weighted averaging mechanism to generate a binary classification (authentic or synthetic) with a confidence score. The system achieves 97.8% accuracy on the Fake-or-Real dataset, 96.4% on ASVspoof 2019 LA (EER 1.45%), and 94.2% on ASVspoof 2021 LA (EER 3.82%), with a processing time of 0.12 seconds per audio file. A continuous learning module enables periodic model retraining with automated data augmentation to adapt to evolving deepfake techniques. The invention is applicable to banking authentication, telecommunications security, digital forensics, and voice-based biometric systems.

Disclaimer: Curated by HT Syndication.