What question did this study set out to answer?

This research aims to enhance the accuracy of detecting deepfake audio using machine learning techniques.

March 21, 2026Open Access

Deepfake Audio Detection Via MFCC Using Machine Learning

Puntos clave

This research aims to enhance the accuracy of detecting deepfake audio using machine learning techniques.
Literature review conducted on existing deepfake audio detection systems.
Mel-frequency cepstral coefficients (MFCCs) used as input features.
VGG16-based convolutional neural network (CNN) employed for initial classification.
Hybrid CNN-LSTM models utilized to capture both spatial and temporal features.
Hybrid models improve accuracy in classifying real and fake audio.
Overcoming limitations of traditional CNNs in learning time-based dependencies.

Resumen

The emergence of AI-generated voices has posed significant problems with the authenticity of media and their digital safety. False audio detection or fake audio has been critical in such areas as audio forensics and voice authentication. In this paper, a literature review of deep fake audio detection with deep learning is conducted. The system used currently works with Mel-frequency Cepstral Coefficients (MFCCs) as the input feature and a VGG16based Convolutional Neural Network (CNN) as transfer learning to classify the real and fake voices. VGG16 is an effective model that can capture spectral variations but it is not able to learn temporal dependencies. To overcome this hybrid CNN-LSTM models have been investigated, which combine both spatial and time based feature learning to make them more accurate and robust.

Leer artículo completoexternamente

Me gusta

Guardar

Ver artículo completo