Neural Network Strategies and Models for Voice Cloning in a Multi-speaker Mode: An Overview

Abstract

The evolution of data science and the constant challenge of carrying out different processes using a few resources with simultaneous personalization has promoted interest in the development of voice cloning. Nowadays, different machine learning techniques are used, given their efficiency in generating relationships across multiple parameters. In this regard, we evaluated the best-performing models and the different process optimization strategies within this sector, where through neural network models separated modularly by their functionality, it is possible to generate independent processes taking into account the most significant number of linguistic factors in the generation of the voice, thus obtaining significant results of a clear improvement in the whole process of synthesizing the voice of a target speaker.