Voice Spoofing Detection Using Long Short-Term Memory Models with Mel-Spectrogram Analysis
IAN BAEK, Sojung Min · Journal of Student Research · 2025
Voice spoofing prevention has become a primary concern due to the increasing trend of frauds and scams using machine learning technology. To address this issue, I propose a novel voice spoofing detection system that classifies voices with accuracy, determining whether they are genuine or artificially generated. This system uses the fourier transform to convert the voice input into a mel-spectrogram, which is then processed using convolutional neural networks and long short term memory networks for classification. Comprehensive experiments demonstrate that the proposed method effectively classifies input signals with significant accuracy. In addition, I conducted frequency masking experiments to study how specific frequency bandwidths are correlated with enhancing real-fake classification. I expect that this method will inspire further innovations to prevent voice spoofing attacks.