Research proposal

["Albert Bregman, Tel: 514-398-6103" <IN09@xxxxxxxxxxxxxxxx>]

                                        September 20, 1992

Dear AUDITORY members:

I am interested in doing a study on the relation between speech
and primitive auditory organization.  It is related to the claim
by Al Liberman and some of his colleagues at Haskins laboratories
that speech removes from the incoming acoustic data whatever it
needs to form the words and then leaves the remainder to be heard
as non-speech sounds.

Let me describe the plan for the experiment, and then ask you
some questions about it.

THE PLAN:

My idea for the study is to use sine-wave analog speech.  This
is speech in which each formant is replaced by a single
sinusoidal tone whose frequency follows the same pattern over
time as the formant did in the original speech.  Such analogs can
be heard as speech but this interpretation is not necessarily
compelling.

I plan to use two kinds of biasing with the subjects:  Some
subjects would be encouraged to hear them as speech, and others
to hear them as "computer" sounds.  Liberman's hypothesis would
be that the subjects in the "speech" group would require more
intensity to hear out one of the component tones.

To summarize, then, there would be a form of biasing, followed
by a test.  Here are more details:

(a) Biasing:

    Group 1 (Speech bias).  The listeners are practiced on the
     analogs.  Each is alternated with the word it mimics.
     Listeners rate the analogs on how easy they are to
     understand.

    Group 2. (Non-speech bias) The listeners are told that these
     "complex tones" are being considered as computer generated
     warning signals for the work-place and they are to rate them
     on some global qualities. e.g.,  how pleasant, smooth, or
     attention-getting they are.  They are alternated with non-
     speech sounds for comparison.  "Which one sounds smoother
     (or more pleasant, or more attention-getting)?"  Listeners
     in this group would be given different degrees of exposure
     to the analogs to see whether familiarity alone had an
     effect on hearing the parts inside them.  They would also
     get some sort of encouragement to hear them as coherent
     sounds (I haven't figured out how yet.)

     N.B.  Group 2 can't simply be taught to hear the analogs as
     sets of whistles, because this training would make the tones
     audible for them at a lower level in the test, making it
     appear that Group 1's relatively worse performance
     (requiring greater intensity) was due to the fact that the
     interpretation as speech had made it harder for Group 1 to
     hear the tones.


 (b) Test procedure:

On each trial, one of the analogs is alternated one of the tones
that forms part of it.  Subjects adjust the loudness of the tone
inside the word until it can barely be heard.

The outcome that would favor Liberman's position would be if
Group 1 required greater intensity to hear the embedded sinusoid.


MY QUESTIONS:

Do you think that this is a fair test of the theory?

Are the methods likely to work?

Would the results be susceptible to other interpretations?

Do you know of any facts that bear on how the results would come
out?

FINALLY:
     Do you think that the debugging of experiments or other
     projects beforehand is a good way to use the AUDITORY list?
     One wouldn't have to worry about others stealing one's ideas
     since the ideas would have been put forward in a public way
     on the list, associated with the name of the originator.
     Furthermore, it would be possible to avoid unnecessary
     duplication of efforts (people would tell you about similar
     studies in progress) and it might stimulate collaborations.
     What do you think?

I would encourage people to answer to the list as a whole.

 - Al Bregman