Speech Synthesis from neural decoding of spoken sentences