I've now finished going through all Mark's stuff, and I think the
following comments are positive and/or constructive:
1. The reiterant speech is great - well done!
2. The formant tracking problem is serious, but it seems that most of the
errors are confined to particular contexts:
(a) where the voice is breathy,
(b) in certain segments/syllables, primarily /r w l nasals/.
We can dismiss the /l nasals/ set as too much for any tracker, and
/r/ is often a problem too. So /w/ is the only serious disappointment.
3. We hope the following suggestions can be discussed at the meeting:
(a) Mark tries to speak in a louder voice. Our observation of his
intentional contrastive stress is that the voice is rather more modal
in these sentences, and so the tracker tends to be more accurate.
(b) We check whether there is a lower frequency limit set on each
formant. If so, it is set too high for Mark's /w r/s, and we can look into
whether we can change this limit - initial excursions into ESPS haven;t
been helpful, and Mark probably has a lot more knowledge about this than
we do. But I'm happy to continue to find out. (We can of course change
various default settings, but Ali's efforts with Mark's voice so far
merely confirm my experience, which is that getting the perfect settings
for one utterance only introduces new problems in another: the entropics
defaults seem to be pretty good.)
(c) We discuss speaking style a little, particularly with respect to
stress, because (pace Jill) Ali and I still feel that Mark's stress
contrasts are fairly extreme. (I list a few deatils below).
In terms of making a final decision, Ali and I would be happy to use
Mark's voice if we can reduce the amount of breathy voice to a level that
the formant tracker is reasonably successful, and we feel that his talking
a bit louder overall would probably do the trick.
Hope that sounds ok.
Stress comments: Ali and Sarah's perception (we're happy to be found to be
in the minority).
Set 1 - nuclear stress intended to be noncontrastive
Sets 2 and 3 - intended to be contrastive.
Our impressions:
Set 1: contrastive stress (or at least rather high drama) in 1d thru 1h
(to run it, it;s wonderful, the idea of it, for YorkTalk, to stare at).
Sets 2 and 3: all clearly contrastive stress, but we hear it as on the
very constrastive side.
We do think this is a relatively minor issue, but we also think we should
try to agree on thetype of speech style wanted. Some may depend on whether
Mark thinks he normally speaks this way, or whether it is a function of
reading/recording. If he normally speaks this way, then presumably we
should let him do it!
looking forward to seeing you all on Monday
Sarah
______________________________________________________________________
Dr. Sarah Hawkins Email: sh110@cam.ac.uk
Dept. of Linguistics Phone: +44 1223 33 50 52
University of Cambridge Fax: +44 1223 33 50 53
Sidgwick Avenue or +44 1223 33 50 62
Cambridge CB3 9DA
United Kingdom