Hi once again,
I have now recorded 100 prompts, and still my EHMM voice is failing to build.
The unit selection voice with the same database builds fine.
The commands I execute are (in this order):
./bin/do_build build_prompts
./bin/do_build label
./bin/do_build build_utts
./bin/do_clustergen f0
./bin/do_clustergen mcep
./bin/do_clustergen voicing
./bin/do_clustergen combine_coeffs_v
./bin/do_clustergen generate_statenames
./bin/do_clustergen trajectory_ola
./bin/do_clustergen dur
Is this correct? I am using Festival 2.1, Festvox 2.4, and
speech_tools-2.0.96-beta. Should I be invoking a different chain of commands?
Or is the problem indeed that I have too limited phonetic coverage in my
database?
If I need to provide information about what is printed at a particular stage, I
would be more than happy to do so if someone would be able to offer some
suggestions on what might be going wrong.
Basically what I am trying to achieve is a voice that is more even, not as
choppy and bouncy as a unit selection voice but not as static and dull as a
diphone voice. Is EHMM the right technique to reach this middleground?
Kind regards,
Philip Bennefall
_______________________________________________
Festlang-talk mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/festlang-talk
