[From: Bruce Nevin (Wed 93105 12:07:54 EDT)]
( Avery Andrews 941005.1753 ) --
Here is an experimental setup that I have had in mind. (I have described
it here before.) There is a card that enables you to modify a speech
signal in real time, with some C programming on the card. Martin is
familiar with the card. I don't recall the details just now.
I would like to set this up so that the auditory signal heard by a
participant through headphones is modified in real time. For example, I
would like to shift the formants so that the E of "bed" and "better"
sounds as though the speaker is saying "bid" and "bidder", and so on for
other words. I would feed this speech signal back through headphones to
the speaker in real time. She or he will also hear the voices of other
participants (i.e., me), unmodified, through the headphones. We are to
carry on a conversation whose subject matter is ostensively the purpose
of the exercise. If she is controlling the auditory perceptual input,
she will start saying "bad" and "badder/batter" and so on for other words
with this vowel E. If she is concurrently controlling the kinesthetic
perceptual input (tongue position), she may also modify her pronunciation
of "bad" etc., and even of other sounds that are normally kept distinct
("bah!", "father", "baud", "bother, etc.). I hope that it will be
possible to tease out what is being controlled. I anticipate that
for vowels it is primarily auditory perceptions that are controlled, and
that for consonants kinesthetic and touch perceptions become more
important. This makes for a fundamental dichotomy between the two classes
of speech sounds, vowels and consonants, and could give real insight into
perceptual control of syllables and of what we think of as stress, meter,
etc.
Rick Marken described some odd effects of playing certain kinds of noise
in headphones, in which the vowel quality seemed to be shifted in context
of the noise. It might be possible to use this phenomenon in an
analogous way, though probably with less control of the variables
involved. I say probably, because it might be possible to determine the
effect with some precision and then do the experiment. I had hoped to
convince someone in the speech synthesis group at BBN to help me set this
up. However, I no longer have access to people at BBN in Cambridge since
my job changed last Fall.
These are things I plan to undertake when I have the resources to do so.
Just now I'm saving up to replace the hard disk and controller on my
somewhat antique 286 clone, and that's about my speed for the immediate
future.
I am enjoying reading things on the list, though I haven't time to read
everything, and certainly not to respond.
Bruce