[hnjst]

There is a generalization allowing asymptotically optimal entropy extraction from Markov chains traces that doesn't require transition weights knowledge (like Von Neumann trick doesn't require bias knowledge).

https://ieeexplore.ieee.org/document/1684974

Shameless plug of a paper on related topics: https://link.springer.com/article/10.1007/s00446-017-0311-5