Voice cloning is a highly preferred feature for personalized speech interfaces. Speaker coding is based on training a separate model to directly assume a new speaker embedding from cloning audios and to be used with a multi-speaker propagative model.