VALL-E can be used to synthesize high-quality personalized speech with only a three-second enrollment recording of a speaker as an acoustic prompt. The model of the voice can then be used for text-to-speech applications.