[hts-users:03405] CMU-ARCTIC-SLT demo
During the generation of global variance in the Training.pl script of
CMU-ARCTIC-SLT demo, we try to separate silence labels and generate a new
model without these redundant information. make_data_gv() subroutine
undertakes this job by generating a list of silences under the file:
What I could not understand is that the script only generates model for
silences in the beginning of utterance and silence contexts for the end of
the utterances is not included in the list. I wonder if there is a
particular reason for this situation. In the synthesis results using my
own database, I have observed some explosive distortion at the end of
utterances and I think there may be a connection between the two.
Thanks in advance.
- [hts-users:03406] Re: CMU-ARCTIC-SLT demo, Keiichiro Oura