Detection of Deepfake Audio Using Long Shortterm Memory Networks Comparing With Multilayer Perceptron's for Improved Accuracy

This research is to improve deepfake audio detection accuracy by employing two comparative analysis of deep learning Algorithms one is Long Short-Term Memory Networks and Multilayer Perceptron's. A higher level of precision compared to conventional methods is the aim. Methods and Materials: To evaluate the prediction of audio authentication, a dataset comprising unique variables was analysed and divided into two groups, Group 1 and Group 2. Numerous elements are covered by these variables, including voice characteristics, spectral bandwidth and acoustic. They used the LSTM technique in the execution phase and compared it to MLP networks. 20 % was decided aside for testing and the remaining 80 % for training With G-power settings a value for alpha is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 5}$</tex> and an average <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{G}$</tex>-power of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 5 \%}$</tex>. Result: Utilizing Group 1 and Group 2 results, SPSS analysis carried out <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 4. 3 0 \%}$</tex> accuracy for Long Short-Term Memory Networks Group 1's enhanced accuracy is better than Group 2's, Multilayer Perceptron's accuracy of 84.60 %. The statistical analysis confirmed the significance of the higher accuracy noted in Group 1. A separate sample T-test was used to support this, and the results showed a p-value of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 0 4 0}$</tex>, which indicates statistical significance (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{p}&lt;\mathbf{0. 0 5}$</tex>). The most efficient method for detecting audio authentication is believed to be the suggested strategy. Conclusion: In conclusion, our research demonstrates that Long Short-Term Memory Networks outperform Multilayer Perceptron's in the identification of false sounds. The utilization of LSTM-based models enhances precision and fortifies defences against the ever-evolving techniques of audio manipulation.