Contributions of Jitter and Shimmer in the Voice for Fake Audio Detection

Fake audio detection (FAD) aims to identify fraudulent speech generated through advanced speech-synthesis techniques. Most current FAD methods rely solely on a deep neural network (DNN) framework with either waveforms or commonly used acoustic features extract high-level representations, overlooking the analysis of prosody differences between genuine and fake speech. Prosody carries important cues about naturalness emotional content, which can be leveraged in audio. This paper explicitly investigates information represented by jitter shimmer features. On basis our investigation, we found strong evidence that obvious exist level real speech, particularly feature has large dynamic variation for To ensure accurate estimation F 0 better propose using two additional methods, YIN SWIPE, place IRAPT algorithm extraction process. Moreover, design DNN-FAD system combining Mel-spectrogram The effectiveness proposed method is evaluated datasets Audio Deep Synthesis Detection (ADD) 2022 2023 challenges. experimental results show both static continuous features, especially extracted SWIPE algorithms, provide complementary knowledge traditional spectrum-based systems. optimal effectively reduce equal error rate from 41.29 % 35.77 ADD2023 challenge, achieving relative improvement 13.37 %.