Frequency Domain-Based Detection of Generated Audio

Attackers may manipulate audio with the intent of presenting falsified reports, changing an opinion a public figure, and winning influence power. The prevalence inauthentic multimedia continues to rise, so it is imperative develop set tools that determines legitimacy media. We present method analyzes signals determine whether they contain real human voices or fake (i.e., generated by neural acoustic waveform models). Instead analyzing directly, proposed approach converts into spectrogram images displaying frequency, intensity, temporal content evaluates them Convolutional Neural Network (CNN). Trained on both genuine voice synthesized signals, we show our achieves high accuracy this classification task.