Automatic Video Dubbing

The continuous development of artificial intelligence affects the development of all spheres of life, including the media space. Major video hosting companies are now trying to solve the problem of translating video from one language to another through an intelligent system. The realization of such a concept is one of the possible ways to increase the target audience of video platform users several times. Due to the constant development of deep learning, the solution to this problem can be presented as a variant system consisting of several neural network models of different architecture and tends to be constantly improving and searching for new strategies that can eliminate the problems encountered and the shortcomings of already existing approaches. This work is aimed at implementing a primitive system of automatic video dubbing, analyzing modern approaches to solving the problem and highlighting significant shortcomings of the basic implementation, revealed in the process of implementation on real videos. Special attention is paid to the study and search for rational solutions to the problems discovered at the stage of system application. These are such problems as: separation of extraneous noise and speech, correlation of the original and duplicated audio tracks, as well as matching the lip movement of a talking person with spoken vocals, i.e. lip synchronization. During experiments, both statistical and neural network methods and approaches are used to solve the set tasks.