Research on Voice Cloning with a Few Samples

Abstract

With the rapid development of computer technology, voice technology has become a research hotspot in the field of deep learning. Allowing computers to listen, see, speak and feel is the future development direction of human-computer interaction. Among them, voice becomes the most promising human-computer interaction method in the future which has more advantages than other interaction methods. Voice cloning is one of the branches of voice technology, which can imitate the voice of a specific person. In order to solve the problem of the need to provide a large number of samples and long waiting time for voice cloning in the past, a method to complete real time voice cloning with only a few samples is proposed. This method is different from the traditional model. Three models are used for joint modeling, and different databases are used for independent training. The vocoder uses a new type of LPCNET, which can achieve good results on some samples and low-performance devices.