Comparison of the Use of the DEMUCS Neural Network On Different Platforms for the Separation of Sources Of Musical Origin

Abstract

This paper makes a comparison between 3 systems deployed on different platforms (Web, Desktop, Mobile) which implement the DEMUCS neural network, responsible for separating sources of musical origin. The objective of this work is to determine on which platform the neural network can be executed more quickly for the use of the average user and from this to propose an optimal architecture for standard development. For this purpose, we selected 12 songs to be separated in the systems of the 3 platforms mentioned and we measured the time it takes for each system to execute the required separation and thus choose the best platform as a starting point. The results and conclusions of the work support the reason for choosing the platform, from which the development architecture was proposed.