[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:03259] Is there an inconsistency of HMgeTool with logF0 stream?


I am using HMgeTool of HTS 2.2 to perform MGE training for F0
parameter. I found that HGen.c::CreatePdfStreams creates storages for
pst->mseq and pst->vseq with genInfo->tframe components (including
both voiced and unvoiced frames), whereas HGen.c::SetupPdfStreams only
uses pst->T components of these storage (usually pst->T <
genInfo->tframe for logF0 stream). Is this an redundancy (or
inconsistency) ?

Moreover, it seems that HMgeTool has no mechanism to deal with the
possible voiced/unvoiced mismatches between original and generated
speech. Is it right?

Can anyone help me with these questions? Thanks in advance!


Ninh K. Duy