[hts-users:01201] evaluation

["Sarah Creer" <smcreer@xxxxxxxxxxxxxx>]

Hello,
I've been trying to find a way to evaluate the voices that I've been
building objectively. Although I want to use human listeners later on,
I'd ideally like a way to do objective evaluation to track the
progress and see if there is a way to measure the improvement, for
example when using different amounts of adaptation data.
I was hoping to get an overview of what other people have been using
to evaluate their voices objectively, either using each feature
(spectral info, log f0, duration etc.) separately or looking at the
output as a whole. I am currently looking into finding distances
between speech synthesised by my models and original target speaker
data and have looked at papers using these techniques but if there are
any other techniques that anyone uses I'd really like to hear about
them.
Many thanks for your help,
Sarah Creer


------------------------------------------------
Sarah Creer
PhD student
Clinical Applications of Speech Technology group
Department of Computer Science
University of Sheffield
UK
smcreer@xxxxxxxxxxxxxx