Mel spectrogram-based audio forgery detection using CNN

In this time of technology, digital speech can be created and falsified by a very diverse hardware software technologies. Audio copy-move forgery is an audio technique that goals to create forged hiding undesirable words or repeating wanted in identical speech. Therefore, authentication has been necessary requisition. study, effective approach spectral images based on detection using convolutional neural networks (CNN) with data augmentation proposed. There are only few handcrafted methods conducted for the forgery. None existing works proposed deep feature learning from recording Mel spectrogram. This first method employ spectrogram The CNN architecture classifies suspicious into two classes: original forged. system successfully trained these image extraction. algorithm tested our datasets generated Arabic Speech Corpus TIMIT database. results show effectiveness, robustness post-processing operations, high accuracy compared other studies.