Re: Alain's note on vowel periods and subtraction

[Alain de Cheveigne <alain@xxxxxxxxxxxxxxxxxxx>]

At 10:56 AM 97.5.26 +0000, Pierre Divenyi wrote:
>If the model were true to life, it would do the job much more efficiently
>for the dynamically changing f0 (and f1 through f3???) case than for
>the steady-state case. If, however, it cannot deal with the problem, or
>deals with it with equal or lesser efficiency than with the steady-state
>vowels, then it would go down the drain, elegance notwithstanding.

>From an "engineering" point of view, changes of F0 or spectrum (F1, etc.)
make the fundamental period somewhat harder to estimate.
The model might thus perform less well when the competing vowel is
modulated (except if F0 modulation introduces F0 differences that
weren't present in the static condition).  Just how much depends on
how sharp the transitions are.  On the other hand, dynamic changes
of the target should make little difference.

I don't see why the model would need to perform better for dynamically
changing F0 or formants.  Experimental data suggest that modulating
the F0 of either target or competitor has little effect on segregation,
as long as it doesn't change the average F0 difference (see for example
Summerfield, Q. (1992). "Roles of harmonicity and coherent frequency
modulation in auditory grouping," in "The auditory processing of speech:
from sounds to words," Edited by M. E. H. Schouten, Berlin, Mouton de
Gruyter, 157-166).

As for formant transitions in V+V or V+CV pairs, Peter Assmann found
in some cases that they improved the identification of the vowel without
the transition.  However that was the case only when F0s were the same.
F0-guided segregation cannot operate in that case, so this result says little
about harmonic cancellation.  When F0s were different (and F0-
guided segregation was effective), a formant transition had no effect
(Assmann, P. F. (1995). "The role of formant transitions in the perception
of concurrent vowels,"  JASA 97, 575-584).

In summary, F0-guided segregation based on harmonic cancellation
(with my model or others) should not be better for dynamically
changing stimuli.  If anything it might be worse.  However dynamic
changes might possibly benefit to mechanisms other than F0-guided
segregation.

Alain

Alain de Cheveigne', alain@linguist.jussieu.fr