VALL-E emerges in-context learning capabilities and can be used to synthesize high-quality personalized speech with only a 3-second sample. VALL-E synthetically preserves speaker's emotion and acoustic environment.