1 paper
Zeyang Song, Tianchi Liu, Tianrui Wang +3
Current TTS systems typically rely on open-loop, single-pass generation and can produce sporadic local prosodic defects, such as misplaced stress, unnatural pauses, or flattened in…