• Version
• Download 6
• File Size 672.48 KB
• File Count 1
• Create Date 18 May 2025
• Last Updated 18 May 2025

Deep Fake Audio Recognition Using Deep Learning

Authors:

Madhuri Borawake^*1, Kiran Raut^*2, Aniket Patil^*3, Karan Shelke^*4, Shivam Yadav^*5

Department of Computer Engineering,

Pune District Education Association’s College of Engineering, Manjari Bk.,

Hadapsar, Pune, Maharashtra, India – 412307

Abstract - Deep fake audio is incredibly lifelike synthetic audio that can be produced because to recent advancements in deep learning algorithms. This poses a major threat to digital communications' legitimacy, security, and privacy. Deep fake audio detection has become a critical challenge since current techniques cannot keep up with the rapid advancements in audio synthesis technology. The objective of this study is to develop a dependable deep fake audio detection system using Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks. The proposed method consistently distinguishes between real and false sounds using two advanced audio feature extraction techniques: spectrograms and mel-frequency cepstral coefficients (MFCCs). The RNN and LSTM-based models are trained and evaluated on a range of datasets of deep fake and actual audio samples to guarantee their effectiveness in practical settings. The importance of deep fake audio detection for privacy protection, maintaining the legitimacy of digital communications, and ensuring the veracity of audio evidence in court is highlighted by this study. By demonstrating how deep learning approaches may be used to counter the growing threat of deep fake audio, the findings pave the way for further advancements in this important subject.

Key Words:   Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN), Mel-frequency cepstral coefficients (MFCCs), Deep Learning