Can monaural temporal masking explain the ongoing precedence effect?

Abstract

The precedence effect for transient sounds has been proposed to be based primarily on monaural processes, manifested by asymmetric temporal masking. This study explored the potential for monaural explanations with longer (“ongoing”) sounds exhibiting the precedence effect. Transient stimuli were single lead-lag noise burst pairs; ongoing stimuli were trains of 63 burst pairs. Unlike with transients, monaural masking data for ongoing sounds showed no advantage for the lead, and are inconsistent with asymmetric audibility as an explanation for ongoing precedence. This result, along with supplementary measurements of interaural time discrimination, suggests different explanations for transient and ongoing precedence.

1. Introduction

The precedence effect is readily observed for both transient sounds and longer “ongoing” sounds with durations greater than about 100 ms. The transient precedence effect refers to temporal asymmetries in the localization and lateralization of brief sounds, just a few milliseconds in duration. With a pair of binaural pulses or noise bursts presented over headphones, for example, the interaural cues of the initial binaural transient contribute more to the lateral position of the composite intracranial image than the second transient (Wallach et al., 1949; Yost and Soderquist, 1984; Shinn-Cunningham et al., 1993). Other studies have demonstrated the precedence effect for ongoing sounds (Zurek, 1980; Braasch et al., 2003; Dizon and Colburn, 2006; Freyman et al., 2010; Donovan et al., 2012; Pastore and Braasch, 2015; Freyman and Zurek, 2017). Freyman et al. (2010) explored the precedence effect for conditions with both abrupt and gradual gating and found only modest differences in the strength of the precedence effect. Such findings demonstrate that the ongoing lead sound determines lateralization to a much greater extent than the ongoing lag sound, without a strong influence of an abrupt onset.

One theoretical picture of the transient precedence effect is that it results from the peripheral response to the lead transient being stronger than that to the lag. Fitzpatrick et al. (1999) measured responses at several points in the ascending auditory system and found clear evidence for a stronger lead response at sites as peripheral as the auditory nerve, with progressively increased lag suppression at more central sites. Hartung and Trahiotis (2001) and Xia and Shinn-Cunningham (2011) were able to predict the transient precedence effect by proposing specifically that hair cell adaptation is the temporally asymmetric mechanism that leads to stronger representation of the leading transient. On a behavioral level, Gaskell and Henning (1999) demonstrated strongly asymmetric monaural masking for pairs of brief sounds, with the lead sound producing more forward masking of the lag sound than the reverse. These behavioral and physiological data combine to support the notion that the transient precedence effect may have a simple explanation in terms of a relative peripheral response, rather than an intrinsic binaural mechanism.

There is no comparably simple model of the ongoing precedence effect. Because the effect is observed with sounds with very slow rise times, and with durations that extend beyond the period of peripheral dynamics, it is unlikely that the mechanism presumed to underlie the transient effect could play a role in the ongoing effect. The notion can be tested behaviorally with a masking study, parallel to Gaskell and Henning (1999), which would assess the strength of peripheral representations of ongoing lead and lag stimuli. Finding a large masking asymmetry, with the lead component more audible than the lag, would support a peripheral-representation mechanism for the ongoing precedence effect; little or no masking asymmetry (or reverse asymmetry) would be evidence against such a mechanism.

The primary goal of this study was to provide those masking data, treating either the lead or the lag component of (monaural) ongoing stimuli as the signal and the other as the masker. These results were compared to similar conditions tested with transients (i.e., one lead-lag burst pair), essentially replicating Gaskell and Henning (1999).

A secondary goal of the study was to compare interaural time delay (ITD) discrimination for the same transient and ongoing signals presented binaurally. Past studies with transients have shown that objective (i.e., forced-choice) measurements of ITD discrimination on the lagging sound are poorer than those on the leading sound, with parameter dependencies like those of lateralization measurements (Zurek, 1980; Shinn-Cunningham et al., 1993). Inclusion of ITD discrimination allowed us to explore such associations between lead/lag masking and binaural sensitivity with both transient and ongoing stimuli.

2. Methods

2.1. Subjects

Two female listeners, S1 and S2, participated in all the conditions of this study. Another subject (S3, male) participated in the monaural masking conditions, while a fourth subject (S4, female) participated in the ITD discrimination conditions. The ages of the subjects ranged from 22 to 28 yrs. All had audiometric thresholds ≤20 dB HL at octave frequencies between 500 and 8000 Hz in both ears. They practiced extensively before data collection began.

2.2. Monaural masking

Stimuli used to study masking for ongoing signals were monaural versions of binaural stimuli used in the past (Freyman et al., 2010) to elicit the ongoing precedence effect, where ITDs alternated between +500 and −500 μs. The stimuli constructed from these bursts are illustrated in Figs. 1(A) and 1(B). Each 1-ms noise burst is shown by a rectangle with a number inside indicating a particular token of noise. For lag detection, the lead bursts only [the unshaded rectangles in Fig. 1(A)] were presented in three of the four temporal intervals on each 4AFC trial. In the fourth, randomly selected interval, the lag bursts (shaded rectangles) were also presented at a specified level below the lead. Lead-lag pairs with different noise samples were presented every 4 ms, as shown by the numbers. The same set of samples was used across the four intervals of a single trial, while a fresh set of noise tokens was used on each trial. The subject's task was to detect the interval that contained the lag bursts. The example shown indicates the ongoing stimulus with 63 burst pairs (approximately 250 ms). To study lag detection for transient sounds, only the first pair of bursts was presented. The stimuli used to investigate lead detection are similar and are shown in Fig. 1(B).

Fig. 1.

Schematic diagrams of noise burst stimuli used in detection and ITD discrimination tasks. Monaural Delay was either 1.5 or 2.5 ms (2.5 ms is shown here), resulting in onset to onset delays between the lag and following lead of 2.5 and 1.5 ms, respectively. The Lag Delay in (C) and (D) is always 2 ms. See text for a detailed description.

The delay between monaural lead and lag components is termed Monaural Delay in Figs. 1(A) and 1(B). Two values of Monaural Delay were tested to mimic the temporal configurations of a particular precedence effect stimulus. As seen below [Figs. 1(C) and 1(D)], the imposition of interaural delays on lead and lag bursts creates different Monaural Delays. With the Lag Delay parameter [from lead onset to lag onset in Figs. 1(C) and 1(D)] equal to 2 ms, and with τ = 500 μs, the Monaural Delay for the left ear in Fig. 1(C) would be 2.5 ms, while it would be 1.5 ms in the left ear for the stimulus in Fig. 1(D). Both of these Monaural Delays were tested in the detection task [the diagram in Figs. 1(A) and 1(B) shows a 2.5-ms Monaural Delay]. The description of the transient stimuli is identical to that of the ongoing stimuli, but with only the first burst pair presented. Stimuli were generated on a personal computer with a 24-bit sound card, low-pass filtered at 8.5 kHz, attenuated, and delivered to the right ear through a TDH 39 headphone (www.telephonics.com) via a headphone amplifier and a passive attenuator at a level of 70 and 84 dBC SPL for the ongoing and transient stimuli, respectively.

The stimuli were presented in blocks of 25 trials where the condition and signal-to-masker ratio (SMR) were fixed, with the first 5 trials serving as practice and the last 20 counted. Correct-answer feedback was provided on every trial. Extensive preliminary testing was conducted with each subject to determine the appropriate SMRs and provide practice. A minimum of four blocks were obtained at each SMR with a descending order of SMR (easiest to hardest) across blocks until all SMRs had been completed once for each condition, before repeating the sequence three more times. Additional smaller blocks were sometimes necessary if results across blocks were not stable. In total, 80 trials (minimum) were tested for each point on the psychometric function for each condition.

2.3. ITD discrimination

The stimuli were again made from 1-ms binaural noise bursts. Either the lead or lag bursts were diotic, while the others had an ITD of τ favoring the left or right ear in one interval and the opposite ear in the other interval. The examples shown in Fig. 1 are for signals with an ITD favoring the right ear imposed on the lag bursts [Fig. 1(C)] and on the lead bursts [Fig. 1(D)]. The Lag Delay, from the onset of the lead burst to the onset of the lag burst, was fixed at 2 ms, and the delay between burst pairs, 4 ms. During a trial the left leading ITD was in the first or second interval with 50% probability. For transient stimuli just one burst pair was presented per interval.

The apparatus was the same as for monaural detection except that the stimuli were delivered through a pair of Etymotic ER2 insert earphones (www.etymotic.com) at a level of 50 dBA for the ongoing trains. No adjustments in the attenuator settings were made for the transient stimuli. The test stimuli were generated and presented at an 80 kHz digital-to-analog conversion rate, permitting 12.5 μs resolution (25 μs resolution in ITD between intervals) when zero padding was used to create the delays. When it became clear that this resolution was not sufficient for all conditions, a method for creating subsample delays was employed. Some subjects repeated a few conditions to ensure that the new and old methods gave similar results; the majority of the data collection used the second method. In no case were lead and lag discrimination measured for the same stimulus condition with a different delay method.

After listening to the two intervals on each trial, one of which had a right-leading ITD and the other left-leading, subjects reported whether the image moved to the left or the right via a key press. For the transient signals, the offset to onset time between the two intervals of a trial was 0.5 s, and for the ongoing stimuli it was 0.3 s. Correct-answer feedback was provided on every trial. A range of ITDs, whose values were based on preliminary data obtained with each subject individually (which also served as practice), was tested using the method of constant stimuli. The ITD was fixed within a block of 20 trials that followed five practice trials in each block. Test ITDs varied from largest to smallest across a series of blocks. Once each ITD required for constructing a psychometric function for both lead and lag had been tested, those ITDs were revisited three more times for a minimum of four blocks (80 counted trials) for each ITD. Additional blocks were sometimes added if results across blocks were not stable.

3. Results

Psychometric functions were plotted and fitted with logistic functions, with interpolated thresholds (62.5% correct for detection and 75% correct for discrimination) estimated. Figures 2 and 3 show the psychometric functions and fits for the four subjects. Not surprisingly, there were some differences in thresholds across subjects. S1, for example, had generally lower detection thresholds than the other subjects, while S2 exhibited better ITD discrimination thresholds. Despite these differences in absolute threshold values, data from all subjects showed similar trends.

Fig. 2.

Results of detection and ITD discrimination with transient stimuli. The points in each panel form psychometric functions that are fitted with logistic functions. Filled and open symbols apply to detection or discrimination of the lead and lag components, respectively. Note the shift in scale between the left and right detection panels to accommodate the different ranges of data for 1.5 and 2.5 ms Monaural Delays. The lag discrimination data from S4 were too flat to be fitted appropriately with a logistic function.

Fig. 3.

As in Fig. 2, with ongoing stimuli.

As expected based on published temporal masking recovery functions (e.g., Gaskell and Henning, 1999), the detection functions for transients (Fig. 2), with Monaural Delays of 2.5 ms show better performance at lower SMRs than the data for 1.5 ms delays. The functions show a consistent and large advantage for lead versus lag detection, indicating more forward than backward masking, in agreement with the results of Gaskell and Henning (1999). In contrast, the results for the corresponding ongoing conditions (Fig. 3) do not show this reduction in lag relative to lead detection. If anything, lag detection was slightly better, especially for S1 and S3. Across subjects and both Monaural Delays for the transient condition, the average threshold was nearly 12 dB better for lead detection than lag detection. For the ongoing condition, the average threshold was 3 dB worse for lead detection than lag detection. Thus, there was no evidence that monaural lag detection was poorer than lead detection for the ongoing stimuli and some evidence of a small effect in the opposite direction.

For ITD discrimination, the psychometric functions in Figs. 2 and 3 show clear evidence of reduced sensitivity to the lag ITD. (Note the different ITD axes in Figs. 2 and 3.) For transients, the difference is striking, with all lag ITD thresholds well over 200 μs and at least six times larger than lead thresholds. Lag ITD with ongoing stimuli was generally much more discriminable, averaging 51 μs, although this was still between two and three times the 18 μs average for lead ITD discrimination. Taken together, the results of the two experiments indicate that a correlate of ongoing precedence can be measured with an objective ITD discrimination task, but that monaural contributions to the reduced binaural sensitivity to the lag are not evident in the monaural masking data.

4. Discussion

The purpose of this study was to evaluate whether explanations for the precedence effect based on decreased monaural response to the lag sound within lead-lag pairs could be ruled out. Results showed significant temporal masking asymmetries for single pairs of brief monaural noise bursts. Forward masking averaged about 12 dB greater than backward masking for Lag Delays of 1.5 and 2.5 ms. These values are comparable to the masking asymmetries observed by Gaskell and Henning (1999), and indicate that the lag burst is at a lower sensation level than the lead burst when the two are presented at equal levels. As Gaskell and Henning (1999) concluded, this masking asymmetry cannot be excluded as a basis for the precedence effect for transient stimuli.

A stronger statement than this would require a quantitative demonstration that compensating for the temporal masking asymmetry offsets the precedence effect. Such data were not collected either by Gaskell and Henning (1999) or in the current study. However, it is noteworthy that Xia and Shinn-Cunningham (2011) observed that the precedence effect was offset by a 12 dB increase in lag level for a stimulus similar to that used here (wideband transient with a 2 ms lag delay, their Fig. 5), matching the average temporal masking asymmetry of 12 dB reported in the current study. Further data collection and analysis of more conditions such as these in a within-subjects design may prove useful in determining the extent to which the precedence effect for transients can be explained by temporal masking.

In contrast, the data obtained with the ongoing stimuli indicated no masking asymmetry in favor of lead bursts relative to lag bursts. Based on this result, the ongoing precedence effect, at least for noise burst stimuli like those used here (Freyman et al., 2010), is unlikely to be accounted for by monaural processes that reduce the neural representation of the lag in relation to the lead. The strong ongoing precedence effects seen in studies using noise that was not modulated (Braasch et al., 2003; Dizon and Colburn, 2006; Pastore and Braasch, 2015) are also not likely to be associated with a masking asymmetry. Lead versus lag detection with continuous noise stimuli are expected to be equal (for the same absolute value of delay) based on the equality of their power spectra. A brief experiment by Ando and Alrutz (1982) supported this expectation by showing no difference in detectability when a noise signal is added with a lead or a lag (of the same absolute value) to its coherent noise masker. Therefore, it appears that masking symmetry applies to both noise-burst trains and continuous noise.

The results of the ITD discrimination experiment with transient stimuli showed that listeners were much less sensitive to the interaural delay of the lag sound relative to that of the lead sound. This relative insensitivity to the lag ITD has been reported previously for brief sounds (Zurek, 1980; Shinn-Cunningham et al., 1993) and is consistent with lateralization being dominated by the lead ITD. The results also showed that sensitivity to the ITD of the lagging bursts in ongoing stimuli was poorer than that of leading bursts, although not as dramatically so as for transients.

When the two experiments are considered together, transient and ongoing precedence effects appear quite different. ITD sensitivity on a lagging transient is substantially decreased relative to the lead, suggesting that binaural representation of the lag in the lead-lag pair is reduced in some way. The monaural masking data suggest that this degraded sensitivity could at least partially result from reduced peripheral response to the lag component. These results fit neatly with the physiological results mentioned in Sec. 1 suggesting that a stronger peripheral response to the lead than to the lag may be a key source of the precedence effect with transients.

Conversely, the ongoing precedence effect is not associated with asymmetric masking for the stimuli tested here. Further, even though transient and ongoing effects have similar strengths in lateralization studies (e.g., Braasch et al., 2003; Shinn-Cunningham et al., 1993; Freyman and Zurek, 2017), the present ITD discrimination results indicate a greater lead/lag asymmetry with transients than with ongoing stimuli.

These results leave the theoretical picture of the ongoing precedence effect murkier than that of the transient effect. With comparable peripheral representation of lead and lag components, at least as measured by masking, and with lag ITD discrimination only moderately worse than lead discrimination, accounting for the ongoing precedence effect will likely involve different mechanisms than those used to explain transient precedence. One possibility that deserves exploration is the enhanced emphasis of ITD on the rising parts of modulations in ongoing signals (Dietz et al., 2014). It is conceivable that a mechanism giving rise to that effect might not be susceptible to asymmetric masking, and so would be consistent with current results. It is also plausible that, with that mechanism, ongoing lead/lag ITD discrimination might be less asymmetric than for transients, as observed here. Tests of these speculations will come only from detailed modeling work.