[hts-users:00784] Flatness of Pitch on HTS

[Carlos Mendes <cmdm@xxxxxxxxxxxxxxx>]

Hi,

Recently I've been trying to use HTS with Portuguese voices. I'm using
the training script used in one of your HTS Demos, and found my results
a little bit curious, mainly the flatness of the pitch during the entire
utterance. I has expecting some pitch smoothing, but not so accented.
The HMM based voices that I've heard so far, have good intonation, not
almost flat as I obtained. I ended up solving this by manually adjust
the pitch mean and standard deviation to compensate the effect( using
HTS arguments), and the result sounded much more natural ( I don't like
this ... looks like a nasty hack ...).
I've also noticed that the pitch is not the only problem ... the segment
durations seam a little bit quasi-constant too... ( I haven't extracted
any information to background this... yet... )

Since I'm new to HMM based synthesis... I guess my questions would have
to be... Is this normal or is it possible that I'm doing something
terribly wrong? and if this is normal can anyone explain me why it happens?

thanks

Carlos Mendes