A Machine Learning Model to Detect Fake Voice

Abstract

Nowadays, there are different digital tools that permit the editing of digital content as audio files and they are easily accessed in mobile devices and personal computers. Audio forgery detection has been one of the main topics in the forensics field, as it is necessary to have reliable evidence in court. These audio recordings that are used as digital evidence may be forged and methods that are able to detect if they have been forged are required as new ways of generation of fake content continue growing. One method to generate fake content is imitation, in which a speaker can imitate another, using signal processing techniques. In this work, a passive forgery detection approach is proposed by manually extracting the entropy features of original and forged audios created using an imitation method and then using a machine learning model with logistic regression to classify the audio recordings. The results showed an accuracy of 0.98 where all forged audios were successfully detected.