The latest advances in computer vision processing need tools for creating realistic deepfakes. Deepfake technology has made it viable to provide very practical-looking fake audio, video, and images. In the recent era, we have faced issues that include impersonation and disseminating false information. The spread of fake media streams causes chaos in social networks and has the potential to harm a person's or a group’s reputation, manipulating public sentiment, as well as mindsets about the individual or community. To address these challenges and detect deep fake video images and voices, a deep learning approach has been adopted in this study by using two publicly available datasets containing audio, videos, and images for research and experiment. Six models such as CNN, LSTM, and Pre-trained models VGG-16, Alex-net, Google-Net, and Mobile-net have been analyzed for deepfake detection. Our proposed CNN approach achieved an accuracy of 98.89% on the face and an accuracy of 98.96% on the voice dataset and LSTM achieved an accuracy of 95.11% on the face and 92.67% on another voice dataset. A comparative performance analysis of the Deepfake analysis results obtained in this study is performed with related works to show the supremacy of the current study.