Detecting Manipulated Audio Using Adaboost Machine Learning Algorithm

The proliferation of deepfake audio necessitates robust detection methods. This work explores the efficacy of Mel- Frequency Cepstral Coefficients (MFCC) features and AdaBoost for deepfake audio identification. This proposed a pipeline involving extraction of MFCC features from segmented audio, followed by training an AdaBoost classifier on a labelled dataset of genuine and deepfake audio samples. The classifier leverages the discriminative power of MFCC features and AdaBoost ensemble learning capabilities to distinguish between authentic and manipulated audio. Among the advantages of this method are its ability to capture speech traits, strong resistance to noise and low computational requirements compared with certain deep learning methods. The recognition in this study is that we must keep updating models continuously and training them with various data sets as long as ever-changing tactics applied by creators of deepfakes continue to be an issue. Additionally, consider other audio functions which can be incorporated into it as well as hybrid methods where deep learning techniques would also be used. They have raised points about ethics concerning these technologies for detecting fake videos and call for their responsible development and use. With a research background, the proposed model described in this paper has achieved 92% accuracy rate; being better than any other models discussed here. In general terms then, what has been done here shows us just how much potential there could be behind MFCC features together with AdaBoost technique when it comes down detecting fake audios but still needs some adjustments here and there so that we apply them responsibly.