Temporal decline of masking and comodulation masking release

Temporal decline of masking and comodulation masking release Dennis McFadden and BeverlyA. Wright . Departments of Psychology and Speech Communication and Institute for Neuroscience, University of Texas, Austin, Texas 78712

(Received 20 May 1991; revised 29 January 1992; accepted 3 April 1992)

Masking sounds can be continuously present, gated simultaneously with the signal, or gated somewhat prior to the signal. This continuum of relative onset times was explored using waveforms of the sort commonly employed in studies of comodulation masking release (CMR). There was a 50-Hz masker band centered on the 1250-Hz tonal signal,, and four 50- Hz flanker bands centered at 850, 1050, 1450, and 1650 Hz. In some conditions, all four flanker bands had the same temporal envelope, and the masker band either had that same envelope (correlated presentations) or a different envelope (uncorrelated presentations). In other conditions, all five bands had different temporal envelopes (all-uncorrelated presentations). The masker band and/or the four flanker bands were either gated nearly simultaneously with the signal (burst conditions) or were gated prior to the signal by a duration that was systematically varied (fringed conditions). The eight listeners could be partitioned into three groups on the basis of their response to these fringing manipulations. Two listeners (the large fringers) showed a gradual improvement in detectability with increasing fringe duration (called a temporal decline of masking), while three others (the small fringers) showed little improvement in detectability. For the remaining three subjects, there was evidence of a "learning" effect that changed them from large fringers to small fringers over a 10-week period of listening. When present, the temporal decline of masking was greater for the correlated than for the uncorrelated comodulation condition; as a consequence, the difference in detectability between them (the comodulation masking release or CMR) increased with fringe duration. By fringing the masker and flanker bands separately and in combination, it was revealed that the temporal declines of masking were primarily attributable to the fringing of the flanker bands. In contrast, large CMRs required long fringes on both the masker and flanker bands. The above results were obtained with 50-ms signals, but generally similar data were obtained with a signal duration of 240 ms. The difficulties raised for experimentalists and theorists by such long-term practice effects are discussed.

PACS numbers: 43.66.Dc, 43.66.Ki, 43.66.Mk

INTRODUCTION

Simultaneous-masking experiments typically employ maskers that are either continuously present or are gated on and off synchronously ("burst") with the signal. Some- times, however, masker and signal are gated off together but the onset of the masker precedes that of the signal by a "fringe" duration that is systematically varied. When the masker is broadband and the signal is both tonal and longer than about 50 ms, there is typically little or no difference in detectability for burst and continuous maskers (for reviews see McFadden and Wright, 1990; and Wier et al., 1977). An exception to this near equivalence of burst and continuous (or long-fringed) maskers can occur when the duration of the tonal signal is shorter than about 30 ms. Then, signal detectability can be as much as 10-20 dB worse when masker and signal onsets coincide than when there is a 100- to 200- ms masker fringe preceding the signal, an effect commonly known as overshoot (e.g., Elliott, 1965, 1967, 1969; Zwicker, 1965a, b). Another exception to the near equiv-

a) Portions of this work were reported at the 115th Meeting of the Acousti- cal Society of America [J. Acoust. Soc. Am. Suppl. 1 83, S34 (1988) ].

alence of burst and continuous maskers exists when the

masker and signal do not overlap spectrally. Then, even for long signals, detectability can be as much as 10-35 dB worse in the burst than in the continuous condition (e.g., Green, 1969; Leshowitz and Cudahy, 1975; Bacon and Viemeister, 1985a, b; McFadden and Wright, 1990), and fringed maskers are often used to explore the transition from the former to the latter. •

Recently, we have reported 20-30 dB improvements in detectability as the fringe duration was increased for certain complex maskers that did not overlap the signal spectrally (McFadden and Wright, 1990). Improvements of this sort have previously been called "adaptation of masking" (e.g., 'Viemeister, 1980), but we prefer the more neutral phrase "temporal decline of masking" because it is descriptive with- out appearing to suggest an explanation of the effect. Tem- poral declines of masking appear to be a common effect whenever signal and masker do not overlap spectrally, but we happened to encounter them in the context of studying a set of phenomena popularly known as comodulation (e.g., Hall et al., 1984). In our previous paper (McFadden and Wright, 1990), we reported on a task known as the comodu-

144 J. Acoust. Soc. Am. 92 (1), July 1992 0001-4966/92/070144-13500.80 @ 1992 Acoustical Society of America 144

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 128.114.34.22 On: Thu, 27 Nov 2014 21:17:11

lation detection difference, or CDD (Cohen and Schubert, 1987; McFadden, 1987; Wright, 1990). In that task, the signal is a narrow band of noise and the masker is one or more

narrow-band noises located in frequency regions flanking the signal region. Here we report the results of similar fringing manipulations in a different comodulation task---comodulation masking release, or CMR (e.g., Hall et al., 1984). In the version of CMR of interest here, a tonal signal is em- bedded in one narrow-band noise (the masker band) and there are one or more other narrow-band noises present (the flanker bands).

I. METHODS

A. Waveforms

The signal was a tone of 1250 Hz obtained from an oscil- lator (General Radio 1310A); its duration was either 50 or 240 ms, in different experiments. The narrow-band noises used for masker and flanker bands were all 50 Hz wide and

were constructed as described in McFadden and Wright (1990). Briefly, digitally synthesized sinusoids were summed to produce waveforms that were stored as computer files, the contents of which were supplied to digital-to- analog converters and low-pass antialiasing filters ( - 27 dB at 6.25 kHz) at the appropriate points in the trial-timing sequence. The spacing of the summed sinusoids was 2 Hz and the sampling rate was 10 kHz. The masker band was centered at 1250 Hz, and its overall level was 67 dB SPL (re: 20/•Pa). When flanker bands were present, there were always four ofthem, centered at 850, 1050, 1450, and 1650 Hz. The flanker bands either all had the same temporal envelope or all had different temporal envelopes. When the flanker bands all had a common envelope, the masker band either had that same envelope (the correlated condition) or a different envelope (the uncorrelated condition), in different blocks of trials. When the flanker bands all had different

envelopes, the masker band was unlike any of the flanker bands (the all-uncorrelated condition). The levels of the individual flanker bands at the different center frequencies were intentionally varied so as to present a variety of spectral "profiles" (Green, 1988). The scheme for scrambling levels involved always presenting one of the four flanker bands at about 67 dB overall and attenuating the other three bands by either 2, 4, 6, or 8 dB, in a pseudorandom pattern. There were 28 such different files of four flanker-band waveforms

(and 28 masker-band waveforms) for the computer to select from on each observation interval, so the waveforms in the two observation intervals of a single trial were the same only by chance. Data were also collected with only the masker band present (masker-only condition).

B. Procedure

Throughout, the psychophysical procedure was adaptive, two-interval forced choice with feedback. Since the primary variable of the experiment w•ts the relative duration of the masker and flanker bands, the observation intervals of the trials ranged from 55-495 ms across blocks of trials, but otherwise, the trial-timing sequence was as follows: warning

interval ( 100 ms), pause (500 ms), first observation interval, pause (500 ms), second observation interval, response interval ( 1000 ms), and feedback interval (350 ms). The warning, signal, and feedback intervals were marked by indi- cator lights. Trials were run in blocks of 50. The two- down/one-up decision rule of Levitt ( 1971 ) was used to esti- mate the signal level necessary for 71% correct decisions. The signal level was changed in 4-dB steps until the second reversal, and in 2-dB steps thereafter. The first four reversals of each block were discarded and the remainder averaged. Blocks were discarded if there were fewer than six usable

reversals, if the standard deviation of the reversals exceeded 6 dB, or if there were fewer than 47 responses. Finally, all blocks yielding estimates of sensitivity 15 dB or more below any other block for that subject in that condition were discarded as statistical aberrations; 74 blocks were discarded for this reason, out of more than 4600 blocks collected. The minimum number of blocks acquired for each subject in each condition was five, but because the subjects were run as a group, it was not uncommon to obtain 7-9 blocks for some subjects for some conditions in the process of acquiring the minimum number for others.

As noted, the variable of primary interest in these experiments was the relative durations and onset times of the flanker bands, masker band, and signal. The true burst condition (in which all waveforms are gated on and off together) was not actually run here. In the conditions we will call burst, the masker and/or flanker bands preceded the signal by 5 ms (others have also used short forward and/or backward fringes with their "burst" conditions; e.g., Viemeister, 1980; Bacon and Viemeister, 1985a). Also, a technical limi- tation prevented our generating truly continuous masker and flanker bands, so a fringe of 445 ms had to suffice for the continuous condition [note that in some instances, even quite long fringes have not produced detectability equal to that obtained with a truly continuous masker (Zwicker, 1965a; Green, 1969; Bacon and Viemeister, 1985b; Bacon and Moore, 1986; Bacon et al., 1988)]. The ofjSets of all waveforms were always synchronous. The durations of fringe used with the 50-ms signals were 5 (burst), 25, 45, 195,295, 395, and 445 ms, and the fringe durations used with the 240-ms signals were 5 (burst), 25, 45, and 255 ms. As noted above, trial duration was confounded with fringe duration in these experiments.

Wilsonics (BSIT) gates were used to present all signal, masker-band, and flanker-band waveforms. Cosine-squared rise-decay times of 10 ms were used throughout. All listening was monotic, over TDH-39 headphones mounted in circum- aural cushions. All subjects were tested simultaneously while seated in individual booths located in large sound- proofed rooms.

Performance in the correlated comodulation conditions

was compared to that in both the uncorrelated and the masker-only conditions. When an uncorrelated condition was used as the reference, the difference will be designated CMR (U-C) and when a masker-only condition was the reference, the designation will be CMR (M-C), which is pat- terned on the terminology suggested by Schooneveldt and Moore (1987).

145 J. Acoust. Soc. Am., Vol. 92, No. 1, July 1992 D. McFadden and B. A. Wright: Decline of masking and CMR 145


C. Subjects

Eight undergraduates were employed for the experiments reported here. Their ages were 19-25. All listeners had hearing within 15 dB of normal at all audiometric test frequencies ranging from 500 to 6000 Hz, as determined with an adjustment procedure (Rudmose ARJ4A audi- ometer). Three of these subjects had had one or more semesters' experience in comodulation experiments (including participation in the CDD experiment of McFadden and Wright, 1990), two had had one or more semesters' experience in other psychoacoustical experiments, and three were serving in their first psychoacoustical experiment. At least four 2-h sessions of training preceded the collection of any of the data presented here.

D. Individual differences and practice effects

In a number of our recent experiments (McFadden and Wright, 1990; Wright and McFadden, 1992a, b; Wright, 199 lb), we have been able to dichotomize our subjects on the basis of marked, systematic, and reliable individual differences in performance. Some nominally normal-hearing listeners (the "large fringers") seem to have a general diffi- culty with the burst condition (or with the short-delay condition in overshoot), while others (the "small fringers" ) do not. For the small fringers, detectability in the burst condition is quite good, and it improves only slightly, if at all, as fringe duration is increased. In contrast, the large fringers are much less sensitive than the small fringers in the burst and short-fringe conditions, they then improve substantially as fringe duration is increased, and, while their temporal- decline-of-masking functions do approach those of the small fringers at long fringe durations, they typically do not reach them.

As a result of the current experiment, and others completed since, we now know that this apparent dichotomy among nominally normal listeners interacts with practice. To be specific, we have observed both short-term and long- term practice effects in temporal decline and overshoot tasks. During the first few days of practice (4-6 h) we may see small improvements in detectability in the long-fringe condition (or long-delay condition in overshoot), but it is rare to see a substantial, or a more protracted, improvement in these generally quite stable listening conditions. It is much more common to see short-term improvements in the burst and short-fringe conditions (or in the short-delay conditions in overshoot), and while these improvements can be sizable, they invariably begin to appear early, and soon give the ap- pearance of having asymptoted. In addition, however, we now report that some subjects who initially appear to be large fringers (or "large overshooters" ), will, with extensive practice, substantially improve their performance in the burst and short-fringe (or short-delay) conditions, and, in the process, will become more similar to those who have been small fringers (or small overshooters) almost from the outset. Other subjects will remain stable and consistent over the course of months of study, either as large fringers or small fringers. No subject has ever moved from appearing to be a small fringer (or small overshooter) to being a large

fringer with a given set of waveforms. As noted, the existence of this long-term practice effect

became obvious to us long after the present experiment had been completed, and as a consequence, the present data were not collected in what we now regard to be the optimal way. Specifically, less attention was paid to verifying long-term stability of performance in the burst and short-fringed conditions than should have been. Fortunately, the most important burst condition was retested near the end of our experiment, and evidence of a long-term improvement was observed for three of our eight subjects. On the basis of their early performance, these three subjects would have been classified as large fringers, but the magnitude of the long- term improvement observed in the burst condition moved them into the small-fringer category. Performance of the other five subjects was essentially the same on test and retest, so we have high confidence in their respective classifications.

These facts obviously complicate the task of presenting the current data. In an attempt to report the results in the most complete and accurate fashion possible here, we will show separately the data for three groups of subjects--the two subjects who were large fringers from the outset and remained such, the three who were small fringers from the outset and remained such, and the three "intermediates" who initially appeared to be large fringers but whose retests revealed them to have become small fringers at some time in the approximately 2.5 months separating test and retest. While this form of data presentation is admittedly more complicated than ordinary, we feel it is worthwhile. As noted, three of the current subjects served in the McFadden and Wright (1990) CDD experiment. One was a small fringer there and remained a small fringer here. The other two were large fringers there; one of these remained a large fringer and one became an intermediate.

Large individual differences have been present in the data of numerous other investigators (e.g., Elliott, 1965; Zwicker, 1965a; Green, 1969; Leshowitz and Cudahy, 1975; Wier et al., 1977; Bacon and Viemeister, 1985b; Bacon et al., 1988; Carlyon, 1989; Schmidt and Zwicker, 1991 ), but, to our knowledge, no one has ever given systematic attention to the interacting factors of individual differences and learning when collecting or presenting temporal-decline or overshoot data (also see the Appendix of Wright, 199 lb). The topics of short-term and long-term learning in auditory tasks have been reviewed and discussed by Watson ( 1980, 1987, 1991 ) and Leek and Watson ( 1984); also see Gibson (1953).

The data shown in Figs. 1 and 5 were collected during the first 6-7 weeks of the experiment, in an interleaved man- ner, and mixed with some related conditions that are not shown. During weeks 7-9, data were collected for the conditions shown in Figs. 2-4, and the 50-ms burst condition was revisited during week 10.

II. RESULTS

A. Masker and flanker bands both fringed: 50-ms signals

For one set of measurements, the masker and four ranker bands were all gated together, and the fringe dura-



tion was varied across blocks of trials (see schematic at lower fight in Fig. 1 ). The results are shown in Fig. 1 for the large-fringer, small-fringer, and intermediate groups of subjects. For the large-fringer group [ panel (a) of Fig. 1 ], performance was similar, and relatively poor, in all of the comodulation conditions involving the burst masker (at the far left of the panel). Also, the burst data collected in mid-Sep- tember (the "early" data) were very similar to those collected in late November (the "late" data), indicating essentially no practice effect. In the uncorrelated comodulation condition (open circles), where the temporal envelope of the masker band centered on the tonal signal was different from the common temporal envelope of all four flanker bands, detectability was relatively poor with the burst presentation, and there was a temporal decline of masking of 4-6 dB when 25-445 ms of fringe was added to the masker and flanker bands. In the all-uncorrelated condition (open triangles), in which the masker band and each of the flanker bands all had

different temporal envelopes, detectability was reasonably similar to that for the uncorrelated condition (compare Hall and Grose, 1990). In marked contrast, in the correlated co-

modulation condition (solid circles), where the envelopes of the masker band and all the flanker bands were the same, detectability improved by about 15 dB as the fringe was increased from 5 to 445 ms. Because of these differential tem-

poral declines of masking for the uncorrelated and correlated conditions, the difference between them [the CMR(U-C)] grew for the large fringers from about 2 dB with the burst presentations to about 10 dB when the longest fringe was used. (Statistical analyses of the various differences in these data are precluded by the small number of subjects involved in the three groups. ) For comparison, the panels of Fig. 1 also contain the masker-only data collected with no flanker bands present (open diamonds); for the large fringers, detectability in this condition was essentially the same whether the masker fringe was 5 or 445 ms, in accord with past findings (Hall and Grose, 1990; for reviews see McFadden and Wright, 1990; Wier et al., 1977). Thus the values of CMR (M-C) for the large fringers were negative for all but the longest fringe durations; the actual values ranged from about -- 11 to d- 5 dB as the fringe duration was increased from 5 to 445 ms.

85

80

75

'-- 70

•- t55

n- 150 o

• 55

ß x 80 z

m 75 o

• 7o

65

60

off

(c)

Large Fringers

all-uncorrelated uncorrelated correlated

masker only : :• i i i i

Intermediates

55

50 -•---// ...... ioo DURATION OF FRINGE (ms)

A

0

400

Small Fringers 80

• // .......... 5 20 30 50 1 O0 200

DURATION OF FRINGE (ms)

,

400

75

7O

65

6O

55

5O

SIGNAL

MASKER '

FLANKERS :

FRINGE

50 ms

FIG. 1. Detectability of a 50-ms tonal signal when both the 50-Hz masker band and the four 50-Hz flanker bands were fringed (see schematic at lower right) for durations of 5, 25, 45, 195, or 445 ms. Results are shown separately for the two subjects categorized as large fringers [panel (a) ], the three small fringers [panel (b) ], and the three intermediates [panel (c) ]. The burst data marked "late" were collected about 10 weeks after the "early" data. In the conditions marked correlated and uncorrelated, the temporal envelopes of all four cue bands were identical, and the envelope of the masker band was either the same (correlated) or different (uncorrelated) from that of the common envelope of the cue bands. In the all-uncorrelated condition, the envelopes of all four cue bands, and the masker band, were different. For comparison, data are shown for the masker-only conditions, in which no flanker bands were present. At least five estimates of sensitivity were obtained for each subject in each condition of listening. The vertical flags on the symbols at the top of each panel represent plus and minus one average standard error of the mean obtained by averaging standard errors across all the individual listening conditions and subjects denoted.



Panel (b) of Fig. 1 reveals a similar pattern of results for the small-fringer group, but the magnitudes of the changes were much smaller than for the large fringers. Again, the burst data collected "early" and "late" in the semester were quite similar, as were the data for the uncorrelated and the all-uncorrelated conditions. Here, there was only 1-2 dB of temporal decline of masking for any of the three comodulation conditions, and essentially no temporal decline for the masker-only condition. Thus the values of the CMR (U-C) all ranged from 2 to 4 dB, and the values of the CMR (M-C) ranged from -- 3 to 0 dB. The values of CMR (U-C) were generally larger for the large fringers than for the small fringers, once at least 25 ms of fringe was provided. As was true in the CDD task (McFadden and Wright, 1990), the small fringers both had smaller temporal declines of masking and were generally more sensitive than the large fringers [ compare panels (a) and (b) of Fig. 1 ], meaning that the two groups differed more at short fringe durations than at long.

We wish to argue that the subjects in the intermediate group [ panel (c) of Fig. 1 ], who showed improvement from test to retest with the burst masker and ranker bands, are best thought of as small fringers who simply required extensive practice to overcome their initial difficulties with the burst waveforms. The basis for this belief is that detectability in the burst conditions measured after 10 d- weeks of experience with the fringing task [the "late" data in panel (c) of Fig. 1] was very similar to detectability in the conditions with the longest fringesmjust as was true for the group of subjects who were small fringers from the outset [panel (b) of Fig. 1 ]. This belief is bolstered by our general impression that performance does not change significantly with experience in the long-fringe conditions. It is important to note that had the burst conditions not been revisited, and the practice effect not noted, this group of three intermediate subjects would have appeared to be quite similar to the two large fringers shown in panel (a) of Fig. 1 (although the general sensitivity of the intermediate subjects was closer to that of the small fringers than the large fringers). Also, the CMR (U-C) values for these intermediate subjects would have increased with increasing fringe duration, just as they did for the large fringers.

A summary of the variability in these data, and in the data to follow, is provided in the array of symbols at the top of each panel in Fig. 1, and in subsequent figures. The vertical flags on these symbols indicate plus and minus one average standard error of the mean for the denoted listening condition (function). These mean values were obtained by averaging the within-subject standard errors both across the various values of fringe duration used for that listening condition (across abscissa values) and across the subjects in the subgroup. As can be seen in panels (a) and (b) of Fig. 1, the variability was always greater for the large fringers than for the small fringers, and this relationship held for every listening condition studied in this experiment.

It is worth emphasizing that one's conclusions from the data in Fig. 1 about the effects of fringing on comodulation could be quite different depending upon which of the two measures of CMR was adopted. Below (Sec. III) we argue

that it is preferable to use the uncorrelated (or all-uncorrelated) condition as the reference condition for calculating CMRs rather than the masker-only condition.

B. Flanker bands fringed, masker band constant: 50-ms signals

One might reasonably wonder whether the temporal declines of masking seen in panel (a) of Fig. 1 require fringing of both the masker and ranker bands or whether fringing one or the other would suffice. To answer that question, data were collected with only the ranker bands fringed and the duration of the masker band held constant, or with only the masker band fringed and the duration of the ranker bands held constant.

The data shown in Fig. 2 were obtained by fringing only the four ranker bands; the masker band always began 5 ms prior to the 50-ms tonal signal and ended with it (see schematic at lower right of Fig. 2). The early and late burst data and the masker-only data are reproduced from Fig. 1.

As can be seen in panel (a) of Fig. 2, detectability did improve for the large fringers as the duration of the fringe on the ranker bands only was increased from 5 to 445 ms. For the uncorrelated and all-uncorrelated conditions, the temporal decline was 3-4 dB larger here than when both masker and ranker bands were fringed by 445 ms [panel (a) of Fig. 1 ]. For the correlated condition, however, the temporal decline was only about 8 dB when the ranker bands only were fringed (Fig. 2), as compared to about 15 dB when both masker and ranker bands were fringed (Fig. 1). For the small fringers [ panel (b) of Fig. 2 ], the functions were essentially flat as the fringe of the ranker bands only was increased, but this comes as no great surprise given that there was essentially no temporal decline of masking for these subjects when both masker and ranker bands were fringed [panel (b) of Fig. 1 ]. Again, the third group of subjects would have been categorized as large fringers had only the early burst data been obtained, but the late data reveal that, with extended experience, they are, in fact, more similar to the small fringers.

It is worth noting that, for both the large and small fringers, the values of CMR (U-C) were uniformly smaller in Fig. 2 than in Fig. 1. This suggests that although fringing of the flanking bands is important for a large temporal decline of masking, fringing of the masker band as well is important for a large CMR. Further support for this notion appears below.

For the data shown in Fig. 2, the fringe duration of the masker band was held constant at 5 ms while the fringe duration of the ranker bands was varied. When, instead, the fringe of the masker band was held constant at 445 ms and the fringe of the ranker bands varied, the data in Fig. 3 were obtained. This increase in the duration of the masker band

produced some irregularities in the data, but the important finding is that, again, increases in the fringe duration of the ranker bands did lead to temporal declines of masking for the large-fringer group [ panel (a) of Fig. 3 ]. Also, the temporal decline was again larger for the correlated (18 dB) than for either the uncorrelated ( 5 dB) or the all-uncorrelat-



o

85

8O

75

70

65

60

55

85

8O

75

7O

65

6O

55

(a)

o8

(c)

Subjects: 1 & 5 Large Fringer•

6 all-uncorreloted o uncorreloted ß correlated

• mocker only

Subjecta: 2, 4 & 7 Intermediates

20 ,30 50 100 200 400


(b) • (• Subjects: 3,6 &8 Small Fringers

AO

5 20 .30 50 100 200 400


SIGNAL

50 ms

HASKER [•

FI-4NKEE$ i- 7// ...... '•l FRINGE

85

8O

75

7O

65

6O

55

5O

FIG. 2. Detectability of a 50-ms tonal signal when only the four 50-Hz flanker bands were fringed by durations of 5, 45, 195, and 445 ms, and the duration of the 50-Hz masker band was fixed at 55 ms (see schematic). Results are shown separately for the two subjects categorized as large fringers [panel (a) ], the three small fringers [panel (b) ], and the three intermediates [panel (c) ]. The early and late burst data and the masker-only data have been replotted from Fig. 1. Average standard errors of the mean (within subjects) are indicated at the top of each panel as in Fig. 1.

85

8O

75

• 7O

ac 60 o

• 55

o 85

;• 80 z

• 75

• 70

65

6O

55

5O

Large Fringers

"all-uncorrelated o uncorrelated ß correlated

o masker only : : : :: i i i i

Subjects: 2, 4 & 7 Intermediates


Small Fringers

85

8O

75

70

65

6O

55

5 20 30 50 100 200 400


SIGNAL

50 ms

MASKER [ Fringe: 445 me

FLANKERS T• I< FRINGE >1

FIG. 3. Detectability of a 50-ms tonal signal when only the four 50-Hz flanker bands were fringed by durations of 5, 45, 195, 395, and 445 ms, and the duration of the 50-Hz masker band was fixed at 495 ms ( see schematic). Results are shown separately for the two subjects categorized as large fringers [ panel (a) ], the three small fringers [panel (b) ], and the three intermediates [panel (c) ]. The masker-only data and the data at a fringe duration of 445 ms have been replotted from Fig. 1. Average standard errors of the mean (within subjects) are indicated at the top of each panel as in Fig. 1.

149 J. Acoust. Sac. Am., Vol. 92, No. 1, July 1992 D. McFadden and B. A. Wright: Decline of masking and CMR 149


ed (6 dB) conditions. Here, the small fringers [panel (b) of Fig. 3 ] also showed temporal declines as the fringe duration of the ranker bands was increased. These declines were

about 4 and 9 dB for the uncorrelated and correlated condi-

tions, respectively. For both large and small fringers, the values of

CMR(U-C) were negative with the 5-ms fringe duration, but, because of the greater temporal decline in the correlated than the uncorrelated condition, the values of CMR(U-C) became positive as fringe duration was increased. We note again how different the effects of fringing on camadulation would appear were only values of CMR(M-C) available. The masker-only data shown in Fig. 3 are reproduced from Fig. 1, as are the data shown for a 445-ms fringe on both the masker and ranker bands.

The intermediate subjects [ panel (c) of Fig. 3 ] appear more similar to the large fringers than to the small fringers. While there is necessarily some uncertainty about the form their data would have taken had these conditions been revis-

ited, these fringe data were obtained during the two-week period immediately preceding the collection of the "late" burst data. 2

C. Masker band fringed, fianker bands constant: 50-ms signals

The data in Figs. 2 and 3 indicate that it was primarily the fringe on the ranker bands that produced the large tem-

poral declines of masking seen in Fig. 1 when both the masker and ranker bands were fringed. To confirm this conclusion, an additional experiment was conducted in which the fringe duration of the masker band was varied while the fringe of the ranker bands was held constant at 445 ms. A schematic of the gating configuration, and the results, are shown in Fig. 4. For simplicity, the early and late burst data have been omitted from these panels, but the masker-only and 445-ms data from Fig. 1 are reshown. Here again, the evidence is that fringing the masker band does not contribute to a temporal decline of masking; indeed, over the middle range of fringe values, detectability for the large-fringer and intermediate groups worsened somewhat relative to the short-masker/long-flankers condition at the left in Fig. 4, thereby driving detectability toward the values obtained with the burst masker and ranker bands (at the left in Fig. 1 ). That is, for all subjects, the fixed, long fringe on the ranker bands was itself sufficient to produce nearly all of the temporal decline that was to be obtained. For comparison, shown at the upper right of these panels are data collected with the fringe of the ranker bands fixed at 5 ms and the fringe of the masker band fixed at 445 ms. The data in Fig. 4 were largely obtained during weeks 8-9 of the 10-week period of data collection.

Figure 4 provides additional evidence about the importance of fringing both the masker and ranker bands to the production of a large camadulation effect. For the large fringers, increases in the duration of the masker fringe from

85

80

75

•, 70

• 65 n- 60 o

ae 55

,,o 85

80

75

70

65

60

55

5O

Large Fringers ß Small Fringers

Ranker (me) • i•mla•=•• _• •••- 5 445

• ß all-uncormlated n o uncorreloted ß ß correlated

e ma=ker on•

(c! • {• • • Subject=: 2. 4 & ? 5 Intermediatea

20 30 50 1 O0 200


400

SIGNAL

$0 ms

OR Fringe: 5 ms • FRINGE

5 20 30 50 100 200 400


85

8O

75

70

65

6O

55

5O

FIG. 4. Detectability of a 50-ms tonal signal when only the 50-Hz masker band was fringed by durations of 5, 25, 45, 195, 295, 395, and 445 ms, and the duration of the four 50-Hz ranker bands was fixed at 495 ms (or at 55 ms, data at far right of panels). Results are shown separately for the two subjects categorized as large fringers [ panel (a) ], the three small fringers [ panel (b) ], and the three intermediates [ panel (c) ]. The masker-only data and the data at a fringe duration of 445 ms have been replotted from Fig. 1. Average standard errors of the mean (within subjects) are indicated at the top of each panel as in Fig. 1.

150 J. Acoust. Sac. Am., Vol. 92, No. 1, July 1992 D. McFadden and B. A. Wright: Decline of masking and CMR 150


5 to 445 ms increased the CMR(U-C) from about -- 1 to 10 dB. While this trend does not appear to be present for the small fringers, for both groups the values of CMR(U-C) moved from negative to positive as the duration of the flanker fringe changed from 5 to 445 ms while the duration of the masker fringe was fixed at 445 ms [ squares and circles, respectively, at far right of Fig. 4(a) and (b) ].

D. Masker and flanker bands both fringed: 240-ms signals

In an attempt to determine whether the results reported above are unique to short signal durations, data were also collected using 240-ms signals. Time limitations restricted data collection primarily to the conditions in which both masker and flanker bands were fringed together (as in Fig. 1 ), and technical limitations prevented our exploring the full range of fringe durations used with the 50-ms signal. Also, the preponderance of these data were collected between weeks 3 and 6 of the 10-week experiment, so nothing is known about possible long-term practice effects. For consistency, however, the data are again presented separately for the same three groups of subjects as in the preceding figures.

In Fig. 5 are shown data for the correlated and uncorrelated conditions in which both the masker and flanker bands

were fringed. As can be seen, the pattern of effects, and the individual differences, were generally similar to those obtained with the 50-ms signal (Fig. 1). Specifically, for the

two large fringers in panel (a) of Fig. 5, detectability again improved as the duration of the fringe was increased, and this temporal decline of masking was again greater for the correlated ( 11.4 dB) than for the uncorrelated (7.6 dB) listening condition. Thus, for these subjects, the CMR (U-C) grew from about 6 to 9 dB as the fringe grew from 5 to 255 ms. The small fringers [panel (b) of Fig. 5] were again generally more sensitive than the large fringers, and, while their temporal declines of masking were much smaller than those of the large fringers, the effects were again greater for the correlated (4.6 dB) than for the uncorrelated (0.0 dB) condition; thus the values of CMR (U-C) increased from about 2 to 6 dB over the range of fringe values tested. Finally, the third group of subjects was again intermediate to the other two, having temporal declines of 8.8 and 5.7 dB in the correlated and uncorrelated conditions, respectively, and increases in the CMR (U-C) from about 4 to 7 dB.

For all three groups of subjects, detectability worsened in the masker-only condition as the masker fringe was increased from 5 to 255 ms, in contradiction with the 50-ms data. Similar decreases in detectability have been observed previously with maskers that do, and do not, overlap the signal (Leshowitz and Cudahy, 1975; Carlyon and Moore, 1986; Bacon and Moore, 1987; but compare Wier et al., 1977; McFadden and Wright, 1990), although the signal duration has typically been much shorter. This decrease in detectability means that the values of CMR (M-C) increased with increases in duration of the masker fringe.

85 {al •} Subjects: • & 5 Large Fringers 8O

o,

7o

n,, •0 o

• 55

,,o 85 '• 80 z

m 75 o

-J 70

65

60

55

o uncorrelated ß correlated

* maeker only

Subject=: 2. 4 & 7 Intermediatea

50 ' ;,' ............. 5 20 30 50 100 200 400


(b) 1• • Subjects: 3, 6 & 8 Small Fringers

85

8O

75

7O

5 20 30 50 100 200 400


SIGNAL

240 ms

FRINGE

55

5O

FIG. 5. Detectability of a 240-ms tonal signal when both the 50-Hz masker band and the four 50-Hz flanker bands were fringed for durations of 5, 25, 45, and 255 ms. Results are shown separately for the two subjects categorized as large fringers [panel (a) ], the three small fringers [panel (b) ], and the three intermediates [panel (c) ]. Also shown are data for the masker-only condition, in which no flanker bands were present. Average standard errors of the mean (within subjects) are indicated at the top of each panel as in Fig. 1.



Thus the same basic outcomes can apparently be obtained with signal durations of 50 or 240 ms, just as was true of the CDD task (McFadden and Wright, 1990). This is a potentially important fact for those inclined to see a similarity between these effects and phenomena such as overshoot (e.g., Elliott, 1965, 1967, 1969; Zwicker, 1965a, b; Fastl, 1976), for which brief signals are obligatory.

III. SUMMARY AND DISCUSSION

To reiterate the present findings, (1) we have shown that gating the masker and flanker bands prior to gating the 50-ms signal can improve detectability, but only in some subjects, and only in some listening conditions. Specifically, some subjects (large fringers) showed temporal declines of masking ranging from 6-15 dB, a second group (small fringers) showed little or no temporal decline, and a third group (intermediates) initially performed like the large fringers, but with additional practice, their temporal declines lessened as detectability in the burst (shortest-fringe) condition improved. The large fringers were, in general, less sensitive than the other subjects even with the longest fringe durations tested. All of these facts are in accord with previous, related findings using a different comodulation task (McFadden and Wright, 1990; Wright and McFadden, 1992b). Also, two of the three subjects who had previously served in McFadden and Wright (1990) were independently and identically categorized as a large and a small fringer there and here, and the third common subject was a large fringer there and became an intermediate here. Thus there is evidence for consistency in performance and in classifica- tion. Elsewhere we will demonstrate that groups of sul•jects having large and small temporal declines all easily satisfy the common criteria for "normal hearing. "3 Moore et al. (1990) and Hall and Grose (1990), among others, have also commented upon large individual differences in comodulation experiments, and other examples of long-term practice effects can be found in Hafter and Carrier (1970), Watson ( 1980, 1987, 1991 ), Fendick and Westheimer (1983), Leek and Watson (1984), Kidd et al. (1986), Neff and Callaghan (1987), Kohlrausch and Jacobi (1989), Carlyon (1989), Wright ( 1990, 1991 b), and Richards ( 1992 ). Also, the large individual differences commonly reported in overshoot (e.g., Zwicker, 1965a; Green, 1969) and the difference in overshoot magnitude observed across different psychophysical methods (Elliott, 1965) may be attributable in part to differential and incomplete learning. It is important to em- phasize that whatever changes took place in the intermediate subjects during the 10-week interval between test and retest, they were not the result of continuing experience with the burst condition itself, for only fringed conditions were tested during that time.

(2) The magnitude of the temporal decline was generally similar for the 50- and the 240-ms signals over a common range of fringe values, just as it was in the CDD task (McFadden and Wright, 1990). As we noted in that CDD paper, we regard this to be a rather surprising outcome for it suggests that, whatever changes occur in the auditory system to make the signal more detectable in the fringed than in

the burst conditions, it is only those changes that occur prior to signal onset that matter. Hall and Grose (1990), using a 400-ms signal, observed no temporal decline of masking in a CMR task (no difference whether a masker band and six correlated flanker bands were all continuous or all burst). The basis for this discrepancy may lie in subject or proce- dural differences. In accord with present findings, Hall and Grose (1990) found no difference in the masker-band-only condition whether it was burst or continuous.

(3) When present, the temporal decline of masking was typically greater in the correlated, than in the uncorrelated, condition, meaning that the magnitude of the CMR (U-C) increased with increasing fringe duration. [Interestingly, there was a similar effect in our CDD data collected with a

240-ms signal (McFadden and Wright, 1990, Fig. 4; Wright and McFadden, 1992b), with the consequence that the magnitude of the CDD diminished with increasing fringe duration. ] The larger values of CMR (U-C) with fringed than with burst presentations helps to explain why the CMRs obtained in this lab have historically been smaller than those obtained elsewhere; we have routinely used burst presentations while other labs have tended to use continuous presentations.

(4) Differential fringing of masker and flanker bands revealed that it was the fringing of the flanker bands that was crucial to the temporal decline of masking. Recent overshoot experiments have also emphasized the importance of spectral regions adjacent to the signal region in the control of differences in detectability for brief signals presented near the beginning, or near the end, of a gated masker (McFad- den, 1989; Carlyon, 1989; Schmidt and Zwicker, 1991; Ba- con, 1990a; Carlyon and White, 1992).

(5) A look across the entire set of data reveals that fringing apparently has differential effects on the temporal decline of masking and on the CMR. Namely, large temporal declines of masking seemingly require only that the fringe on the flanker bands be long (Figs. 1-3 and 5 ), but large values of CMR (U-C) seemingly require that the fringe on both the flanker and masker bands be long (Figs. 1 and 3-5 ). That is, while the temporal decline appears to depend primarily on across-channel effects [item (4) above and Sec. IIIA below], the size of the CMR appears to depend upon both across- and within-channel effects. The data in each panel of Fig. 4 are particularly illustrative of these differential effects. When the flanker bands were short and the masker band

'long (square symbols), sensitivity was poor and the CMR (U-C) was actually negative for both the large and the small fringers. However, when both flanker and masker bands were long (fightmost circles), sensitivity was considerably improved and the CMR (U-C) was large. Further, for the small fringers, the CMR (U-C) was, at best, about 4 dB (Fig. 1 ), and whenever the flanker fringe was longer than 5 ms in Figs. 2-4, the CMR (U-C) jumped to that value. One obvious conclusion is that experimenters desiring large CMRs (Moore et al., 1990) should use masker and flanker bands that are continuous or have long fringes.

(6) When the masker and flanker bands were both burst, the CMR (U-C) was larger for the 50-ms signal (Fig. 1 ) than for the 240-ms signal (Fig. 5 ) for the large fringers,



and it was about the same for the small fringers. For comparison, the CDD was larger for the 240-ms than for the 50- ms signal (McFadden and Wright, 1990), although the ranker bands did have different widths and different fre-

quency spacings in the two experiments. (7) In our view, Figs. 1-5 stand as excellent examples of

why the masker-only condition is a poor choice as a reference when attempting to summarize the results of a comodulation manipulation. The effects of fringe duration in these data are sensibly described by the CMR (U-C), whereas the account one would carry away knowing just the values of the CMR (M-C) would be complicated and unnecessarily confusing. One reason for such complicated and confusing ac- counts is that the absolute level of detectability in the correlated (and uncorrelated and all-uncorrelated) condition can be greatly affected by such factors as the level and frequency of the ranker bands (e.g., McFadden, 1986), whereas the masker-only condition is obviously unaffected by these factors. This means that the CMR(M-C) is subject to large variations in magnitude and sign depending upon the stimulus conditions--variations that are basically irrelevant to the issues and mechanisms of comodulation.

(8) We regard our findings about individual differences and practice effects to be important for all experimenters studying comodulation tasks and other tasks involving fringing or onset effects. In particular, enough subjects should be screened or studied to verify that the results being reported do not hold only for a small fraction of the population. Also, sufficient practice must be provided to reduce concern that the subjects are not still in the early stages of learning. Performance may appear deceptively stable to experimenters obtaining only (say) three adaptive blocks of trials per condition, but such similarity of performance cannot be interpreted as reflecting asymptotic performance.

(9) The existence of such large individual differences in temporal decline of masking in apparently normal-hearing listeners [here, McFadden and Wright ( 1990); Wright ( 199 lb) ] has relevance for, and may eventually undercut, explanations of every-day speech perception and linguistic universals that appeal to such adaptation-like processes (e.g., Summerfield et al., 1984; Bladon, 1987).

A. The difference in temporal decline of masking for CMR and CDD tasks

In a previous paper, we reported temporal declines of masking of 20-30 dB, depending upon the specific waveform combinations used. There, the task was CDD, where no masking components spectrally overlapped the signal (McFadden and Wright, 1990). In the CMR task used here, there was spectral overlap of masker and signal, and the temporal declines of masking were substantially smaller than in those CDD experiments. We see this difference as being in accord with past results from related experiments? For example, Green (1969) obtained considerably smaller temporal declines of masking when his tonal maskers were of the same frequency as his tonal signal than when his maskers were off the frequency of the signal (especially when they were on the high side of the signal). This general outcome has been reported by many others (e.g., Zwicker and Fastl,

153 J. Acoust. Soc. Am., Vol. 92, No. 1, July 1992

1972; Leshowitz and Cudahy, 1975; Fastl, 1979; Bacon and Viemeister, 1985a, b; Kimberley et al., 1989; compare Car- lyon, 1989).

In an attempt to explain the differential effectiveness of on-frequency versus off-frequency maskers, Kimberley et al. (1989) and Bacon et al. ( 1988 ) have recently revitalized an idea that dates back at least to Elliott (1965) (also see Wright, 1991 b; Carlyon and White, 1992); namely, that (at least) two mechanisms underlie performance in detection conditions of the sort of interest here, that the two mechanisms individually produce different magnitudes of temporal decline of masking, and that the two mechanisms have different spectral extents. In the Kimberley et al. formulation of the idea, one mechanism is activated by components in the masker that lie above the signal (their frequency-dependent mechanism) and is capable of producing large temporal declines of masking, and the other mechanism is activated by masker components from anywhere in the spectrum (their frequency-independent mechanism) and is able to produce only relatively small temporal declines. That descriptive explanation can be made a little more concrete if one accepts the idea that masking can occur at a number of sites in the auditory system (McFadden and Wright, 1990), and if one presumes that the magnitude and characteristics of the temporal decline seen in a particular listening condition will be determined in part by the primary site of the masking. For example, whenever the masker spectrally overlaps the signal (as in the CMR situation), or lies slightly above or below it in frequency and is relatively intense, the masker is clearly able to contribute directly to the level of activity in the peripheral filter centered on the signal (a within-channel masking effect). However, when the masker lies well above the signal frequency (as in the CDD condition), the preponderance of the masking presumably occurs at a site beyond the (asymmetrical) peripheral filters (an across- channel masking effect). [ In some circumstances, of course, maskers will presumably contribute to more than one site of masking (e.g., the array of noise bands sometimes used in CMR and CDD experiments, and the wideband masker commonly used in overshoot experiments). ] A comparison of our previous CDD data (Mcfadden and Wright, 1990) with the present CMR data suggests that the potential for a large temporal decline of masking is greater when the masking originates from an across-channel interaction than from a within-channel effect, in accord with the Kimberley et al. (1989) formulation. • According to Kimberley et al. (1989), the contribution of the across-channel (frequency- dependent) mechanism can be diminished or lost when there is a sensorineural impairment of hearing (also see Champlin and Mcfadden, 1989; Mcfadden and Champlin, 1990; Ba- con, 1990b). Were that mechanism also to be more subject to individual variation, it might account for the present finding of an apparently bimodal distribution in the magnitude of the temporal decline of masking across subjects in both the CMR and CDD tasks (Mcfadden and Wright, 1990).

B. Adaptation of suppression

Viemeister and Bacon (1982) offered the ingenious sug- gestion (and some supporting data) that temporal declines

D. McFadden and B. A. Wright: Decline of masking and CMR 153


of masking may result not from an adaptation in some exci- tatory process, but from an adaptation of the suppression that is known to exist between adjacent frequency regions. If so, it is unlikely to be the two-tone suppression that is measured psychophysically (Houtgast, 1972), for that shows no evidence of declining as masker/suppressor duration is increased; in fact, psychophysical suppression tends to increase as the common duration of masker and suppressor is extended (Shannon, 1975; Weber and Green, 1978).

C. Implications for research practice

The existence of these long-term "learning" effects raises numerous difficult issues of interpretation and practice that affect both experimentalists and theorists. Perhaps the primary problem is that these long-term effects raise doubts about whether the individual differences so common-

ly observed in these tasks are really as large as they appear, or whether, with enough practice, all large fringers (or large overshooters) will eventually become small fringers (or small overshooters).6 Whatever the ultimate answer to this question, there is the immediate problem of how the experimentalist is to proceed in the interim. How much practice is enough before collection of data for publication begins? Are large fringers to be regarded as a truly different population of listeners that is worthy of serious study and comparison with small fringers, or should they be ignored•r even screened out of our experiments•n the grounds that they are simply slow to learn? Exactly what aspects of a listening condition determine whether it is subject to these long-term effects (see Watson, 1980, 1987, 1991 ) ?

Related to these issues is the seemingly unresolvable question of whether these intermediate listeners really "learned" anything through the long course of exposure to these stimuli; that is, whether there were incremental or quantal changes in some neural structures that then allowed the listeners to better process the signal (or, depending upon the vocabulary one prefers, to better use a listening strategy that previously could not be effectively employed)? A logi- cally possible alternative to true learning is that one day the intermediate listeners simply happened onto a new listening strategy, a strategy that was fully available to them from the outset, but which, for whatever reasons, was not previously attempted, or was prematurely rejected. ? To us, this second alternative is potentially more frustrating than true learning in that the progress of the listeners is seemingly subject more to caprice than to any actions taken by the experimenter. Were there true learning, even if painfully slow, it presumably would proceed in accord with some knowable rules of acquisition, on which the experimenter ought to be able to capitalize. If true learning is occurring in our intermediate listeners, the nature and locus of the structural changes in the nervous system are obviously of great interest, as are such questions as why some listeners have the requisite hardware at the outset of the experiment and others must acquire it during the experiment, why some listeners do not acquire this new hardware even after months of listening, what, if anything, the absence of this hardware means to the every-

day auditory abilities of these listeners, how well this learning generalizes to related listening conditions, and how well it is retained.

When deciding how to behave in the face of these long- term effects, the experimentalist must consider exactly what it is that he or she wishes to know. Specifically, is the concern with what the listener can ultimately do, or with what the listener can currently do? The goal of sensory psychology has traditionally been seen as documenting the limits of sensitivity in various tasks. This suggests that it is the ultimate performance of each listener that is of interest, but achieving that goal, at least for some listening conditions, now seemingly requires that experimenters know in advance whether, or how much of, a difference across listeners reflects a true individual difference in (ultimate) ability and how much can be eliminated by practice of a particular duration. Given these difficulties, an approach other than the traditional one may become necessary. In a different context, Massaro (1978) has argued that what listeners are currently able to do can be more relevant than what they ultimately may be able to do, the argument being that normal-hearing adult listeners have all had generally similar histories of auditory experience and any differences in perceptual ability are therefore interesting reflections of current differences in neural structure or listening strategy. In accord with this argument, some psychoacousticians may decide to ignore possible long-term learning effects and concentrate on what listeners are currently able to do; for example, by practicing on, and then collecting data on, relatively small sets of conditions in relatively short periods of time.

Whatever procedures particular experimenters decide to adopt in response to the problems of individual differences and long-term learning when the psychophysical task or the waveforms are complex, in the future we all should be explic- it about the amount, and type, of practice provided to our listeners for each stimulus configuration for which data are presented. We are hopeful that authors will no longer use comments to the effect that the subjects "were highly experi- enced at psychoacoustic tasks" to justify the collection of only 2-3 blocks of trials per condition with only minimal practice, for, as the current data reveal, similarity of performance across a few early blocks simply cannot be interpreted as reflecting asymptotic performance. General guidelines to the length of practice necessary for various tasks are available from Watson ( 1980, 1991 ). Also, when the tasks or waveforms are complex, we believe it is important for experimenters to test enough subjects so that any common, large individual differences can be detected.

ACKNOWLEDGMENTS

This work was supported by a Research Grant from the NIDCD awarded to the first author, and by the Lloyd A. Jeffress Memorial Fellowship awarded to the second author by the UT Psychology Department. We thank E.G. Pasanen for technical assistance, and Joy B. Runnels for preparing the final figures. We greatly appreciate the helpful comments made by reviewers S. P. Bacon and R. P. Carlyon.



• Detectability can also be substantially different with burst and continuous maskers when the signal/masker configuration is one that leads to masking-level differences (e.g., McFadden, 1966), but the current discussion and experiments will focus on monaural listening exclusively.

2 While the right-most points in panels (a) and (c) of Fig. 3 ( and also of Fig. 4 below) may appear somewhat aberrant, the standard errors of the mean for those points are generally smaller than the average values shown at the top of each panel.

3 Other examples of apparently normal-heating subjects showing dichoto- mous performance have been reported by Jeffress and McFadden ( 1971 ), McFadden et al. (1973), Spiegel and Watson (1984), Leek (1987), and Thibodeau ( 1991 ), and relations between various auditory abilities have been studied by Festen and Plomp ( 1981 ) and Watson (1987). Elsewhere we shall demonstrate that the magnitude of a temporal decline of masking can vary greatly depending upon the reference condition used for its calculation. Specifically, a burst reference condition can produce a larger temporal decline than a "short-delay" reference condition, in which the signal begins with the masker but the masker outlasts it by several tens of milliseconds (Wright, 1991 a, b). Some external evidence suggests that within-channel effects played only a minor role in the production of the CMRs observed in the present experiments. Specifically, Schooneveldt and Moore ( 1989a, b) argued that within-channel influences on the CMR are weakest when the flanker bands are

placed symmetrically about the masker band (as they were here) and when signal duration is less than 100 ms (also true in the majority of the present listening conditions). Watson and Czerwinski (1990) have preliminary data indicating that in Watson's multitone uncertainty experiments the individual differences present prior to the protracted learning are still present, if somewhat diminished, after the learning. Neff and Callaghan (1987) also found individual differences to be "resistant to training." Leek and Watson ( 1984, Fig. 4) have reported some precipitous improvements in performance that are, at least superficially, in accord with this possibility of strategy sampling. In the present experiments the improvements necessarily appear to be quantal because the relevant conditions were only revisited once, but more recently we have seen at least one in- stance of a substantial improvement apparently overnight (Wright, 1991b).

Bacon, S. P. (1990a). "On the frequency regions governing overshoot," J. Acoust. Soc. Am. Suppl. 1 88, S49.

Bacon, S. P. (1990b). "Overshoot in subjects with sensorineural heating impairment," J. Acoust. Soc. Am. Suppl. 1 88, S50.

Bacon, S. P., and Moore, B.C. J. (1986). "Temporal effects in simultaneous pure-tone masking: effects of signal frequency, masker/signal frequency ratio, and masker level," Hear. Res. 23, 257-266.

Bacon, S. P., and Moore, B.C. J. (1987). "Transient masking and the temporal course of simultaneous tone-on-tone masking," J. Acoust. Soc. Am. 81, 1073-1077.

Bacon, S. P., and Viemeister, N. F. (1985a). "Simultaneous masking by gated and continuous sinusoidal maskers," J. Acoust. Soc. Am. 78, 1220- 1230.

Bacon, S. P., and Viemeister, N. F. (1985b). "The temporal course of simultaneous tone-on-tone masking," J. Acoust. Soc. Am. 78, 1231-1235.

Bacon, S. P., Hedrick, M. S., and Grantham, D. W. (1988). "Temporal effects in simultaneous pure-tone masking in subjects with high-frequency sensorineural hearing loss," Audiology 27, 313-323.

Bladon, A. (1987). "Extending the search for a psychophysical basis for dynamic phonetic patterns," in The Psychophysics of Speech Perception, edited by M. E. H. Schouten (Martinus Nijhoff, Dordrecht, The Nether- lands), pp. 258-263.

Carlyon, R. P. (1989). "Changes in the masked thresholds of brief tones produced by prior bursts of noise," Hear. Res. 41, 223-236.

Carlyon, R. P., and Moore, B.C. J. (1986). "Continuous versus gated ped- estals and the 'severe departure' from Weber's law," J. Acoust. Soc. Am. 79, 453-460.

Carlyon, R. P., and White, L. J. (1992). "Effect of signal frequency and masker level on the frequency regions responsible for the overshoot effect," J. Acoust. Soc. Am. 91, 1034-1041.

Champlin, C. A., and McFadden, D. (1989). "Reductions in overshoot following intense sound exposures," J. Acoust. Soc. Am. 85, 2005-2011.

Cohen, M. F., and Schubert, E. D. (1987). "The effect of cross-spectrum correlation on the detectability of a noise band," J. Acoust. Soc. Am. 81, 721-723.

Elliott, L. L. (1965). "Changes in the simultaneous masked threshold of brief tones," J. Acoust. Soc. Am. 38, 738-746.

Elliott, L. L. (1967). "Development of auditory narrow-band frequency contours," J. Acoust. Soc. Am. 42, 143-153.

Elliott, L. L. (1969). "Masking of tones before, during, and after brief silent periods in noise," J. Acoust. Soc. Am. 45, 1277-1279.

Fastl, H. (1976). "Temporal masking effects: I. Broad band noise masker," Acustica 35, 287-302.

Fastl, H. (1979). "Temporal masking effects: III. Pure tone masker," Acustica 43, 282-294.

Fendick, M., and Westheimer, G. (1983). "Effects of practice and the sepa- ration of test targets on foveal and peripheral stereoacuity," Vision Res. 23, 145-150.

Festen, J. M., and Plomp, R. (1981). "Relations between auditory functions in normal heating," J. Acoust. Soc. Am. 70, 356-369.

Gibson, E. J. (1953). "Improvement in perceptual judgments as a function of controlled practice or training," Psychol. Bull. 50, 401-431.

Green, D. M. (1969). "Masking with continuous and pulsed sinusoids," J. Acoust. Soc. Am. 46, 939-946.

Green, D. M. (1988). Profile Analysis: Auditory Intensity Discrimination (Oxford U.P., New York).

Hafter, E. R., and Cartier, S.C. (1970). "Masking-level differences obtained with a pulsed tonal masker," J. Acoust. Soc. Am. 47, 1041-1047.

Hall, J. W., and Grose, J. H. (1990). "Comodulation masking release and auditory grouping," J. Acoust. Soc. Am. 88, 119-125.

Hall, J. W., Haggard, M.P., and Fernandes, M. A. (1984). "Detection in noise by spectro-temporal pattern analysis," J. Acoust. Soc. Am. 76, 50- 56.

Houtgast, T. (1972). "Psychophysical evidence for lateral inhibition in heating," J. Acoust. Soc. Am. 51, 1885-1894.

Jeffress, L. A., and McFadden, D. (1971). "Differences of interaural phase and level in detection and lateralization," J. Acoust. Soc. Am. 49, 1169- 1179.

Kidd, G., Jr., Mason, C. R., and Green, D. M. (1986). "Auditory profile analysis of irregular sound spectra," J. Acoust. Soc. Am. 79, 1045-1053.

Kimberley, B. P., Nelson, D. A., and Bacon, S. P. (1989). "Temporal overshoot in simultaneous-masked psychophysical tuning curves from normal and hearing-impaired listeners," J. Acoust. Soc. Am. 85, 1660-1665.

Kohlrausch, A., and Jacobi, G. (1989). "Learning effects in a simple masking experiment using complex-tone maskers," Abstracts of the 12th Mid- winter Research Meeting, Association for Research in Otolaryngology, 258-259.

Leek, M. R. (1987). "Directed attention in complex sound perception," in Auditory Processing of Complex Sounds, edited by W. A. Yost and C. S. Watson (Erlbaum, Hillsdale, NJ), pp. 278-288.

Leek, M. R., and Watson, C. S. (1984). "Learning to detect auditory pattern components," J. Acoust. Soc. Am. 76, 1037-1044.

Leshowitz, B., and Cudahy, E. (1975). "Masking patterns for continuous and gated sinusoids," J. Acoust. Soc. Am. 58, 235-242.

Levitt, H. (1971). "Transformed up-down methods in psychoacoustics," J. Acoust. Soc. Am. 49, 467-477.

Massaro, D. (1978). Personal communication. McFadden, D. (1966). "Masking-level differences with continuous and

with burst masking noise," J. Acoust. Soc. Am. 40, 141 4-1419. McFadden, D. (1986). "Comodulation masking release: Effects of varying

the level, duration, and time delay of the cue band," J. Acoust. Soc. Am. 80, 1658-1667.

McFadden, D. (1987). "Comodulation detection differences using noise- band signals," J. Acoust. Soc. Am. 81, 1519-1527.

McFadden, D. (1989). "Spectral differences in the ability of temporal gaps to reset the mechanisms underlying overshoot." J. Acoust. Soc. Am. 85, 254-261.

McFadden, D., and Champlin, C. A. (1990). "Reductions in overshoot during aspirin use," J. Acoust. Soc. Am. 87, 2634-2642.

McFadden, D., and Wright, B. A. (1990). "Temporal decline of masking and comodulation detection differences," J. Acoust. Soc. Am. 88, 711- 724.

McFadden, D., Jeffress, L. A., and Russell, W. E. (1973). "Individual differences in sensitivity to interaural differences in time and level," Percept. Motor Skills 37, 755-761.

Moore, B.C. J., Hall, J. W., III, Grose, J. H., and Schooneveldt, G. P. (1990). "Some factors affecting the magnitude of comodulation masking release," J. Acoust. Soc. Am. 88, 1694-1702.

Neff, D. L., and Callaghan, B. P. (1987). "Simultaneous masking by small



numbers of sinusoids under conditions of uncertainty," in Auditory Pro- cessing of Complex Sounds, edited by W. A. Yost and C. S. Watson (Erl- baum, Hillsdale, NJ), pp. 37-46.

Richards, V. M. (1992). "The detectability of a tone added to narrow bands of equal-energy noise," J. Acoust. Soc. Am. 91, 3424-3435.

Schmidt, S., and Zwicker, E. (1991). "The effect of masker spectral asym- metry on overshoot in simultaneous masking," J. Acoust. Soc. Am. 89, 1324-1330.

Schooneveldt, G. P., and Moore, B.C. J. (1987). "Comodulation masking release (CMR): Effects of signal frequency, flanking-band frequency, masker bandwidth, flanking-band level, and monotic versus dichotic presentation of the flanking band," J. Acoust. Soc. Am. 82, 1944-1956.

Schooneveldt, G. P., and Moore, B.C. J. (1989a). "Comodulation masking release (CMR) as a function of masker bandwidth, modulator bandwidth, and signal duration," J. Acoust. Soc. Am. 85, 273-281.

Schooneveldt, G. P., and Moore, B.C. J. (1989b). "Comodulation masking release for various monaural and binaural combinations of the signal, on- frequency, and flanking bands," J. Acoust. Soc. Am. 85, 262-272.

Shannon, R. V. (1975). "Suppression of forward masking," unpublished doctoral dissertation, University of California, San Diego, 149 pp.

Spiegel, M. F., and Watson, C. S. (1984). "Performance on frequency-discrimination tasks by musicians and nonmusicians," J. Acoust. Soc. Am. 76, 1690-1695.

Summerfield, Q., Haggard, M.P., Foster, J., and Gray, S. (1984). "Perceiv- ing vowels from uniform spectra: Phonetic exploration of an auditory after-effect," Percept. Psychophys. 35, 203-213.

Thibodeau, L. M. (1991). Personal communication. Viemeister, N. F. (1980). "Adaptation of masking," in Psychophysical,

Physiological, and Behavioral Studies in Hearing, edited by G. van den Brink and F. A. Bilsen (Delft U.P., Delft, The Netherlands), pp. 190- 199.

Viemeister, N. F., and Bacon, S. P. (1982). "Forward masking by enhanced components in harmonic complexes," J. Acoust. Soc. Am. 71, 1502- 1507.

Watson, C. S. (1980). "Time course of auditory perceptual learning," Ann. Otol. Rhinol. Laryngol. 89, 96-102, Suppl. 74.

Watson, C. S. (1987). "Uncertainty, informational masking, and the capac-

ity of immediate auditory memory," in Auditory Processing of Complex Sounds, edited by W. A. Yost and C. 5. Watson (Erlbaum, Hillsdale, NJ), pp. 267-277.

Watson, C. 5. (1991). "Auditory perceptual learning and the cochlear im- plant," Am. J. Otolaryngol. 12, 73-79 ($uppl).

Watson, C. $., and Czerwinski, M. (1990). Personal communication. Weber, D. L., and Green, D. M. (1978). "Temporal factors and suppres-

sion effects in backward and forward masking," J. Acoust. $oc. Am. 64, 1392-1399.

Wier, C. C., Green, D. M., Hafter, E. R., and Burkhardt, 5. (1977). "Detec- tion of a tone burst in continuous- and gated noise maskers; defects (sic) of signal frequency, duration, and masker level," J. Acoust. $oc. Am. 61, 1298-1300.

Wright, B. A. (1990). "Comodulation detection differences with multiple signal bands," J. Acoust. Soc. Am. 87, 293-303.

Wright, B. A. (1991a). "Temporal effects obtained with signals of various bandwidths presented in notched and unnotched maskers," J. Acoust. Soc. Am. Suppl. 1 89, S 1914.

Wright, B. A. (1991b). "Contributions of stimulus bandwidth, the reference condition, and individual differences to the auditory temporal effect: Evidence for within- and across-channel mechanisms," unpublished doctoral dissertation, University of Texas, p. 277.

Wright, B. A., and McFadden, D. (1988). "Comodulation masking release with delayed signal onsets," J. Acoust. Soc. Am. Suppl. 1 83, S34.

Wright, B. A., and McFadden, D. (1992a). "Evidence that adaptation of suppression cannot account for auditory enhancement or enhanced forward masking," Philos. Trans. R. Soc. London Ser. B (in press).

Wright, B. A., and McFadden, D. (1992b). "Two comodulation tasks and the temporal decline of masking as affected by background noise," in preparation.

Zwicker, E. (1965a). "Temporal effects in simultaneous masking by white- noise bursts," J. Acoust. Soc. Am. 37, 653-663.

Zwicker, E. (1965b). "Temporal effects in simultaneous masking and loud- ness," J. Acoust. Soc. Am. 38, 132-141.

Zwicker, E., and Fastl, H. (1972). "On the development of the critical band," J. Acoust. Soc. Am. 52, 699-702.



Documents

Temporal decline of masking and comodulation masking release