Re: By any other name...

["Richard M. Warren" <rmwarren@xxxxxxx>]

The discussions of illusory continuity observed when gaps in a signal 
have been filled with a louder sound that were sparked by my March 21 
email have been fascinating, and I’ll give some of my comments and 
observations for whatever they are worth.  Bruno Repp’s “philosophical 
(or methodological?) problem” of whether or not the signal is present 
in the noise-filled gap has been thoroughly vetted, so I have nothing 
to add to that discussion.

Yoshitaka Nakajima’s report of the illusory continuity of the longer of 
two crossing frequency glides of different durations that shared “a 
short silent gap (40 ms or less) at their crossing point” is 
interesting.  The gap of 40 ms is much briefer than the 200 or 300 ms 
usually employed in continuity experiments.  Perhaps the broadband 
spectral splatter accompanying the rapid offset of the tone glides may 
have facilitated the illusion.  However, this does not address why it 
is the longer rather than the shorter glide that is perceived as 
continuous.

The effects of gap duration are important.  At gap durations greater 
than 100 ms the higher amplitude sound must be capable of masking the 
fainter for illusory continuity to be heard.  Gaps in a steady-state 
tone can be spanned when filled by a louder broadband noise for up to 
about 300 ms [Warren, Obusek, and Ackroff, 1972, Auditory induction:  
Perceptual synthesis of absent sounds, Science, 164, 586-587].  
Noise-filled gaps up to about 500 ms can be spanned for a tone glide 
[Dannenbring, 1976, Canadian Journal of Psychology, 30, 99-114].  There 
is a striking difference in the continuity limit for tone and for 
noise:  A fainter narrow band noise (1/3-octave centered at 1-kHz) 
interrupted by a louder broader band noise (0.5-2 kHz) can maintain 
illusory continuity for durations as long as 50 seconds [Warren et al., 
1972].

The term “auditory continuity” used to describe the illusion can be 
misleading, since it limits the illusion to complete closure of gaps in 
the signal.  However, it has been reported [Warren et al., 1994, 
Auditory induction:  Reciprocal changes in alternating sounds.  
Perception & Psychophysics, 55, 313-322] that, when contiguous pure 
tones having different frequencies and amplitudes are alternated, 
conditions can be chosen to produce a continuum of illusory 
lengthening, with closure as an easily perceived endpoint.  The extent 
of illusory lengthening is accompanied by a proportional illusory 
decrease in apparent amplitude of the louder inducing tone.  For this 
(and other reasons) I prefer the more general term “induction” to 
“continuity.”

Al Bregman in his email of 22 March considered the perception of 
auditory induction/continuity to be an example of an “old-plus-new 
heuristic” which he spelled out as follows:

      "Whenever a spectrum changes, so as to become more complex or 
louder, especially if the   change is sudden, the auditory system 
should determine whether this changed spectrum could consist of a 
continuation of an old sound with the addition of a new one.  If so, it 
should hear it that way.  The properties of the new sound can be 
determined by calculating what would have had to be added to the old 
sound to obtain the changed spectrum."

This is similar, but represents a somewhat different approach than my 
consideration that auditory induction/continuity represents the 
reversal of masking that employs a contextually determined reallocation 
of a portion of the neural representation of the louder sound for the 
perceptual synthesis of the fainter.  This difference in approach has 
led to Al’s hypothesis that phonemic restoration represents a 
“schema-driven stream segregation” that follows his gestalt principles 
for auditory organization and does not involve reallocation of a 
portion of the interrupting noise (Bregman, Auditory Scene Analysis, 
MIT Press, 1990).  However, Warren et al. (1994) demonstrated that the 
loudness of the interrupting sound does decrease when phonemic 
restoration occurs, as with other types of illusory continuity.

By studying illusory continuity of dynamic signals such as tone glides 
and speech (phonemic restorations), it is possible to examine and to 
discover the special rules underlying the perceptual organization of 
these sounds, and a considerable literature has developed for both tone 
glides and speech.  However, if we shift our attention to the higher 
amplitude sound inducing continuity, some interesting (and puzzling) 
phenomena are observed.  This is best studied using the simplest 
condition:  alternating levels of the same sound.  The information 
gained using this “homophonic” continuity can inform us about the 
neurophysiological coding of amplitude level (i.e., loudness).  Since 
this research is not widely known, I’ll describe it briefly.  Take, for 
example, the case of alternating levels of a 1-kHz pure tone, each 
lasting 200 ms.  When the higher level tone is fixed at 70 dB, and the 
lower (apparently continuous) level is raised in a step-wise fashion, 
the apparent loudness of the higher (70 dB) level decreases (as 
measured by matching with a comparison tone).  When the level 
difference is one or two dB a dramatic change occurs:  Illusory 
continuity and reallocation cease, and the veridical perception of 
alternating levels of the same pure tone is heard by all.  In addition, 
when the amplitude difference approaches the alternation threshold, 
listeners may hear the 70 dB level as a pulsed atonal jangling sound 
superimposed on the continuous tone.

Noise behaves differently.  When alternating levels of broadband noise 
are substituted for the 1-kHz tones, the effects observed for the 
higher amplitude sound differ, indicating differences in the mechanisms 
for loudness coding.  As the apparently continuous lower level noise 
approaches that of the pulsing fixed level noise, the pulsing noise 
appears to become fainter.  But listeners never hear two alternating 
levels of broadband noise.  Even when the difference in the 200 ms 
noise levels is less than 1 dB, the higher amplitude noise is heard 
(paradoxically) as a very faint pulsed noise superimposed upon a much 
louder continuous noise.

Jim Bashford and I are mulling over some speculations concerning 
mechanisms.  These are mainly based on the general principle that 
illusory continuity involves reallocation of a portion of the neural 
representation of the higher amplitude interrupting sound to synthesize 
the missing portions of the fainter sound.  This would leave a 
diminished residue corresponding to the higher amplitude sound.  Does 
the atonal residue that can be heard with alternating tones correspond 
to the perception of neural coding that represents an increase above 
the continuous level?  Why does homophonic continuity with noise differ 
from that with tone?  Any suggestions concerning the reciprocal 
interaction of two levels of the same sound would be appreciated.