Microsoft’s FastSpeech text-to-speech model generates speech up to 38 times faster than the baseline without sacrificing quality.