flâneur

Teaching flow-matching text-to-speech models with RL | Decagon

decagon.ai · saved by 1 readers

The goal of post-training TTS models is not to just achieve a better average sample, but a tighter distribution.

saved by