Fine-Tuning Text-to-Speech Diffusion Models Using Reinforcement Learning with Human Feedback

Anonymous authors*,

Demo Audios

Text Ground Truth Base Model DLPO DPOK KLinR DDPO RWR
"than in the same operations with ugly ones."
"produced the block books, which were the immediate predecessors of the true printed book."
"Caslon's type is clear and neat, and fairly well designed."
"This experiment was so far successful that about 1850 Messrs. Miller and Richard of Edinburgh."
"and these somewhat wiry letters are suitable for the machine process, which would not do justice to letters of more generous design."
"Italy is contentedly stagnant.|Italy is contentedly stagnant."
"America has produced a good many showy books, the typography, paper, and illustrations of which are, however, all wrong."
"oddity rather than rational beauty and meaning being apparently the thing sought for both in the letters and the illustrations."
" it is obvious that legibility is the first thing to be aimed at in the forms of the letters."
" instead of ending in the sharp and clear stroke of Jenson's letters."
" and the elegance and legibility of the ancient more striking than in the Arabic numerals."