[hts-users:02724] Question about the MAP estimation in adaptation
Hi, hts-users & developers
In the adapt demo script, MAP estimation is used after the estimation of linear transforms. When doing MAP adaptation, how are the linear transforms used? In the MAP adaptation chapter of HTK book, it is said that the combination of LR and MAP is to take the linear transformed SI model as prior and to modify them using adaptation data.
However, in this paper:"Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm"[Yamagishi09'], said that the MAP modification take the SI model as prior and the linear transformed adaption data as observations and durations.
How exactly does it work? Thanks in advance.
--
Na Xingyu
Beijing Institute of Technology