This material presents the solution to generate talking face images with the use of deep learning. We conduct the research of existing literature to compose more efficient network design. The final version has additional pre-trained discriminator network to reach superior lip synchronization performance with adversarial training to improve the visual quality of images.
We provide comparative analysis and ablation studies which show insights on how different components of the solution affect the result. This approach achieves comparable consistency in lip movements to other solutions in the field, but has
higher visual quality.