FastSpeech2
FastSpeech2 copied to clipboard
How about adding a discriminator to the Fastspeech2 to improve the naturalness of the spectrum?