Three psychophysical metrics of auditory temporal integration in macaques

Abstract

The relationship between sound duration and detection threshold has long been thought to reflect temporal integration. Reports of species differences in this relationship are equivocal: some meta-analyses report no species differences, whereas others report substantial differences, particularly between humans and their close phylogenetic relatives, macaques. This renders translational work in macaques problematic. To reevaluate this difference, tone detection performance was measured in macaques using a go/no-go reaction time (RT) task at various tone durations and in the presence of broadband noise (BBN). Detection thresholds, RTs, and the dynamic range (DR) of the psychometric function decreased as the tone duration increased. The threshold by duration trends suggest macaques integrate at a similar rate to humans. The RT trends also resemble human data and are the first reported in animals. Whereas the BBN did not affect how the threshold or RT changed with the duration, it substantially reduced the DR at short durations. A probabilistic Poisson model replicated the effects of duration on threshold and DR and required integration from multiple simulated auditory nerve fibers to explain the performance at shorter durations. These data suggest that, contrary to previous studies, macaques are uniquely well-suited to model human temporal integration and form the baseline for future neurophysiological studies.

I. INTRODUCTION

The effect of stimulus duration on psychophysical measures, such as detection threshold, has long been thought to reflect temporal integration or evidence accumulation, a process characterized by a variety of different models across sensory modalities (Gerken et al., 1983; Huk and Shadlen, 2005; Viemeister and Wakefield, 1991; Watson, 1979). Within the auditory domain, the clinical utility of the relationship between duration and detection threshold has been well known for some time, in the form of short-tone audiometry, which uses the detection of short duration tones as a diagnostic measure (Chung and Smith, 1980; Chung, 1981, 1982; Sanders and Honig, 1967). This relationship between the stimulus duration and perceptual measures has also been described experimentally when measuring the reaction time (RT) as a function of the stimulus duration (e.g., Hildreth, 1973), although this is less well characterized in the auditory system. One of the earliest and simplest models of temporal integration is commonly termed “Bloch's law,” which describes a constant time-intensity trade-off between the detection threshold and stimulus duration (Bloch, 1885). An advance in the understanding of the relationship between threshold and duration came in the form of phenomenological models that describe the imperfect integration of intensity over time (Dallos and Olsen, 1964; Gerken et al., 1990; Plomp and Bouman, 1959). Two such models predominate in the psychoacoustics of temporal integration: the power law function and the exponential. Innovation in curve-fitting techniques and the specific form of each function used are crucial considerations and have been reviewed previously (O'Connor et al., 1999). Both the power law and exponential function permit the description of imperfect integration through a rate parameter (a measure of the rate of change of the detection threshold with the duration) and describe integration as only proceeding up to some limited duration. Despite these similarities, the differential ability of these models to accurately describe behavioral data could have implications for the neurophysiological processes underlying the behavior as the work of others has suggested (Gerken et al., 1983; O'Connor et al., 1999). The study of temporal integration has been further advanced by some investigators highlighting the theoretical shortcomings of the exponential and power law models, resulting in the formulation of process models of the effects of the duration, which have been compared to human data (Viemeister and Wakefield, 1991; Meddis and Lecluyse, 2011; Heil et al., 2017).

Animal models provide exceptional utility for studying auditory perception, given the ability to combine invasive electrophysiological techniques and psychophysical measures, and for the use of controlled disruptions to sensory and neural processing. These studies are essential for understanding the anatomical and physiological underpinnings of behaviors which may be explained by temporal integration and described by the aforementioned models, as well as understanding the dysfunction of these mechanisms in models of hearing loss. A variety of species have been used to study these phenomena, including chinchillas, cats, birds, and macaque monkeys (e.g., chinchillas, Clock et al., 1993; cats, Costalupes, 1983; Gerken et al., 1991; Heil and Neubauer, 2003; macaques, Clack 1966; O'Connor et al., 1999; budgerigars, Wong et al., 2019). Much attention has been paid to the rate parameter (the rate of change in detection threshold with duration, described by the exponent m in the power law function and the time constant τ in the exponential function; see Sec. II) and how it differs across studies and species. When a large interspecies data set was analyzed using similar nonlinear curve-fitting methods as in the present study, humans were found to have a mean rate parameter for threshold change (time constant, τ) of about 30 ms, and other species had a median of about 45 ms (O'Connor and Sutter, 2003; O'Connor et al., 1999). This comparison included macaque monkeys (the closest relative to humans that can be used as a subject in such studies), which have been used to investigate temporal integration in just two cases: once using shock avoidance (Clack, 1966) and once using positive reinforcement (O'Connor et al., 1999). Despite this substantial difference in methodology, the two studies found very similar mean time constants in macaques (153 and 141 ms). This is well within the range of all species studied but suggests a rate of temporal integration in macaques that is far slower than that of humans. Moreover, mangabeys, another old-world species, have been found to have time constants of similar magnitude to these previous macaque studies (Brown and Maloney, 1986).

Nonhuman primates, due to their phylogenetic similarity with humans, have long been used as an animal model for human auditory perception (e.g., frequency resolution/selectivity, Gourevitch, 1970; Osmanski et al., 2016; Burton et al., 2018; temporal resolution, Bohlen et al., 2014; stream segregation, Christison-Lagay and Cohen, 2014; speech sound discrimination, detection in noise, Christison-Lagay et al., 2014; Dylla et al., 2013; AM detection, O'Connor et al., 2011; localization, Recanzone et al., 2000; and spatial resolution, Rocchi et al., 2017). And, more recently, there has been a re-establishment of a nonhuman primate model of noise-induced hearing loss (NIHL; Burton et al., 2019; Burton et al., 2020; Hauser et al., 2018; Mackey et al., 2021; Valero et al., 2017) to investigate the histopathological, anatomical, and physiological correlates of perceptual deficits associated with NIHL. However, if behavioral measures of temporal integration in macaques differ substantially from humans as O'Connor et al. (1999) suggested, then this could limit the utility of macaques as translational models of human hearing and hearing impairment. In contrast, the analysis by Heil et al. (2017) suggests that there are substantive limitations in using the exponential and power law models and reported a lack of substantial differences across a range of vertebrate species. To begin resolving these discrepancies, this study reevaluated the effects of tone duration on detection performance in macaques in quiet and noise using both the traditional power law and exponential models and the probabilistic Poisson process model from Heil et al. (2017) to begin investigating the underlying processes.

II. METHODS

A. Subjects

Four adult male rhesus macaques (Macaca mulatta) were enrolled and at the beginning of the study ranged in age from 5 to 10 years old with body weights ranging from 10 to 13 kg. The study lasted about a year, and data collection took about 2–3 months per monkey. Macaques were fed a commercial diet (LabDiet Monkey Diet 5037 and 5050, Purina, St. Louis, MO), supplemented with fresh produce and foraging items. Macaques were also provided manipulanda as well as auditory, visual, and olfactory enrichment on a rotational basis. Macaques were fluid restricted for the study and received filtered municipal water, averaging at least 20 ml/kg of body weight/day (typically closer to 25 ml/kg/day). Their weight was monitored at least weekly (typically 4–5 days each week) and stayed within bounds of the reference weights set to index the animal's health while on study. Macaques were maintained on a 12:12-h light:dark cycle, and all procedures occurred between 8 AM and 6 PM, during their light cycle. After repeated behavioral assessments to attempt to identify compatible social partners, some of these macaques were individually housed (due to incompatibility for social housing with available cohorts) with visual, auditory, and olfactory contact with conspecifics maintained within the housing room. The housing room was located in an American Association for Accreditation of Laboratory Animal Care-accredited facility in accordance with the Guide for the Care and Use of Laboratory Animals, the Public Health Service Policy on Humane Care and Use of Laboratory Animals, and the Animal Welfare Act and Regulations. Macaques in this colony received routine health assessments and tuberculosis testing twice yearly. All research procedures were part of the protocols that were approved by the Institutional Animal Care and Use Committee at Vanderbilt University Medical Center (VUMC).

B. Surgical procedures

Monkeys were prepared for chronic experiments using standard techniques employed in nonhuman primate studies and as reported in previous studies (e.g., Dylla et al., 2013; Bohlen et al., 2014). Briefly, anesthesia was induced via administration of ketamine and midazolam and maintained via isoflurane. A steel headpost was implanted on the skull to restrict head movement during head fixation, minimizing the sound pressure level (SPL) variations at the ear as a result of positioning across behavioral sessions. The headpost was secured to bone using 8 mm titanium screws (Veterinary Orthopedic Implants, St. Augustine, FL) and encapsulated in bone cement (Zimmer Biomet, Warsaw, IN). Multimodal analgesics (pre- and post-procedure), intra-procedure fluids, and antibiotics (intra-procedure) were administered to the monkeys under veterinary oversight. Further details about these procedures can be found in Dylla et al. (2013).

C. Apparatus and stimuli

The behavioral apparatus has been described in detail elsewhere (e.g., Dylla et al., 2013; Rocchi et al., 2017). Monkeys were seated in a primate chair designed and constructed in-house and situated inside a sound treated booth (IAC, model 1200A, Naperville, IL., and Acoustic Systems model ER 247, Arlington Heights, IL.). Stimuli (tones and noise) were presented in the free field via a speaker (Rhyme Acoustics NuScale 216, Middleton, WI ) located in the frontal field at a distance of 90 cm from the center of the monkey's head. The speaker in each booth was calibrated with a ¼ inch microphone (378C01, PCB Piezotronics, Depew, NY) positioned at the location where the monkey's ear canal would be during experiments. The speaker in each booth was calibrated to ensure that outputs were within 3 dB across all frequencies.

The experimental flow was controlled by a computer running OpenEx software (System 3, TDT Inc., Alachua, FL). The tones and noise were generated using a sampling rate of 97.6 kHz. The tone frequencies were 0.5, 1, 2, 2.828, 4, 5.656, 8, 16, 24, and 32 kHz. The tone durations were defined as the time during which the tone envelope was greater than zero. The tone durations were 3.25, 6.5, 12.5, 25, 50, 100, and 200 ms. The linear rise/fall times for the tones were 1.25 ms for 3.25 ms tones, 2.5 ms for 6.5 ms tones, 4 ms for 12.5 and 25 ms tones, and 10 ms for all other tone durations. The noise was broadband (5–40 000 Hz), 76 dB SPL (30 dB SPL spectrum level), and presented continuously from the same loudspeaker as the tones. The monkey pushed and pulled a lever (P3 America, San Diego, CA) to perform the task. The lever state was sampled at 24.4 kHz. More details about the setup can be found in Dylla et al. (2013).

D. Task structure and data collection

Monkeys performed a lever-based RT go/no-go tone detection task. The details of the task have been reported in our previous publications (e.g., Dylla et al., 2013; Rocchi et al., 2017). Briefly, monkeys initiated a trial by pulling a lever. The trials could be signal trials (80%) in which a tone signal of fixed duration was played after a random delay period of 0.8–2.5 s after the lever was pulled, or they could be catch trials (20%) in which no tone signal was played. The monkey was required to release the lever within a response window (600 ms after tone onset) to indicate detection on the signal trials and was required to continue to hold the lever on the catch trials. The lever release on signal trials (hits) were rewarded with fluid. Lack of lever release within 600 ms of the onset of the tone on the signal trials (misses) was taken to indicate non-detection and was not rewarded or punished. The lever release on the catch trials (false alarm) resulted in a 6–10 s timeout in which no trial could be initiated. The correct rejections (lack of release on catch trials) were not rewarded. The experiments were blocked by tone frequency and duration, whereas the tone level was varied randomly trial by trial using the method of constant stimuli. The tone levels in a given experimental block ranged ±30 dB from a user-defined threshold estimate, which could be any level from −10 to 80 dB SPL in 2.5 dB increments. The levels presented in each block were spaced out in a telescoping fashion around the threshold estimate: the seven tone levels in the center were separated by 2.5 dB steps, whereas the two just above and below those were separated by 7.5 dB steps, and the outermost two on the upper and lower edges were 15 dB above/below the nearest level presented (see Fig. 1). Each tone level was repeated 15–20 times, resulting in blocks containing ∼200–265 trials.

FIG. 1.

(Color online) The psychometric functions for all monkeys at 1, 2, 4, and 8 kHz. The different tone durations are indicated by different symbols and colors. The colored curves represent the Weibull fits to the threshold vs level trend. The threshold, indicated by the solid vertical lines, was taken to be the level at a d′ of 1.5. Four of the seven tone durations used in this study are displayed to preserve the visibility of the data.

On a typical day, each monkey would complete ∼1500–2000 trials before reaching satiety, resulting in 7–10 experimental blocks. Consequently, data for a single frequency at all durations tested (7 blocks) were sometimes collected in a single day, but others were split between sessions. The temporal integration experiments were performed after extensive audiometric and noninvasive electrophysiological characterization (e.g., otoscopy, tympanometry, Auditory Brainstem Responses, Distortion Product Otoacoustic Emissions). All of these measures confirmed normal hearing status as reported in our previous work (Bohlen et al., 2014; Dylla et al., 2013; Hauser et al., 2018). These monkeys also had normal frequency selectivity as reported in our previous publications (Burton et al., 2018).

E. Data analysis

1. Calculation of sensitivity (d′) and dynamic range (DR) of the psychometric function

Detection theoretic methods were used to estimate behavioral accuracy (Green and Swets, 1966; Macmillan and Creelman, 2004). The behavioral performance from each block of data was analyzed to calculate the hit rate at each tone level [H(level)] and false alarm rate (F). From these values, the sensitivity was calculated as $d^{\prime }\left(\mathrm{level}\right)=\left(z\left(H\left(\mathrm{level}\right)\right)-z\left(F\right)\right)$, where z represents the conversion to a standard normal variate, an was implemented via the function “norminv” in matlab (2018a; The MathWorks Inc., Natick, MA). The psychometric function was defined as the relationship between d′(level) and the tone SPL. The psychometric function was fit with a modified Weibull cumulative distribution function (CDF) as others have done in both detection and discrimination tasks (Britten et al., 1992; Christison-Lagay et al., 2014; O'Connor et al., 1999; Palmer et al., 2007; also see Dylla et al., 2013 for more on our fitting conventions). The modified equation was ${d^{\prime }}_{\text{fit}}(\text{level})=c-d\ast e^{-({(\text{level}/\lambda )}^{\kappa })},\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\text{for}\hspace{0.17em}\text{level}\ge 0,$ where c represents the saturation, d represents the range of the function, and λ and κ represent the threshold and slope parameters, respectively. Often levels were presented that were below zero and in these cases, the Weibull fit was translated to higher levels before fitting and translated back to the original levels after fitting. The threshold was calculated as that tone level at which ${d^{\prime }}_{\text{fit}}(\text{level})=1.5$ to facilitate comparison with the analysis in O'Connor et al. (1999), which also used this threshold criterion.

The psychometric function slope was calculated by first calculating the range of the tone levels over which ${d^{\prime }}_{\text{fit}}(\text{level})$ spanned (DR), from saturation minus 90% of that range (minimum, min) to saturation minus 10% of that range (maximum, max). This was done by using the c parameter of the Weibull function as an estimate of the saturated performance and d as the range or amplitude of the psychometric function. The slope could then be calculated as (max - min)/DR. However, the reported estimate of the slope is just using the DR measure because when estimating the ceiling performance, d′ is highly variable.

2. Calculation of RT

The RTs were calculated for all of the hit trials and not for the false alarms, misses, or correct rejections. They were calculated as

$$\text{RT}=\text{time}\hspace{0.17em}\text{of}\hspace{0.17em}\text{lever}\hspace{0.17em}\text{release}-\text{time}\hspace{0.17em}\text{of}\hspace{0.17em}\text{tone}\hspace{0.17em}\text{onset}.$$

The RTs were pooled across frequencies at a constant suprathreshold level (35 dB SPL for detection in quiet, 85 dB SPL for detection in noise) to create the distributions shown in Sec. III. The method of constant stimuli used in the current study required that different ranges of levels are presented based on the threshold at a given duration. Because of this shift in the range of the tone levels across durations, it was, at times, not possible to use a single tone level across all durations. In these cases, we used a given level ±5 dB SPL. This made it possible to pool the RTs across the frequency.

3. Statistical analysis and extraction of rate and range parameters

In all of the cases, the curve fits to the Weibull, power law, and exponential functions were attained via nonlinear least squares method implemented in matlab (2018a; The MathWorks Inc., Natick, MA). The Bayesian information criterion (BIC) was calculated in matlab using the nonlinear model fit function, “fitnlm,” which returns multiple information criteria (including the BIC), as well as the goodness of fit measures R-squared and p-values. All time constants (τ, the rate parameter) and constants of the proportionality (I_k is the range parameter) reported were taken from significant fits to the data (p < 0.05). The power law function can be expressed as

$$I_T=I_k{t}^{-m}+I_{\infty }.$$ (1)

The exponential function can be expressed as

$$I_T=I_k\hspace{0.17em}\text{exp}\left(-t/\tau \right)+I_{\infty }.$$ (2)

The statistical analysis of the effects of the tone frequency, noise masker, and duration were conducted using linear mixed effects models (“fitlme”) in matlab (2018a; The MathWorks Inc., Natick, MA). This allowed us to accommodate the data sets with missing points, a common reason for avoiding repeated measures analysis of variance (ANOVA) in cases such as these (Krueger and Tian, 2004). The dependent variable in the models assessing the effects of the tone frequency or noise masker was either τ or I_k, the rate and range parameters, respectively. The background noise, tone frequency, and an interaction term between the two were entered as fixed effects into the model, whereas the intercepts for the individual macaques were entered as random effects. The effects of the duration on the threshold and RT were constructed by entering the tone duration and frequency as fixed effects and the individual monkey as a random intercept term. The detection threshold or psychometric function slope were entered as dependent variables. In all cases, the p-values were obtained by likelihood ratio testing of the model with the effect in question against the model without the effect in question. The T-statistics are reported for each model, similar to the F-statistic often reported for such models, as the F-distribution is equal to the T-distribution squared. The goodness of fit of the Poisson model and exponential function was also assessed by calculating the root mean squared error, which can be calculated as

$$\text{RMSE}=\sqrt{\frac{1}{N}\sum _{i=1}^N{\left(p_i-o_i\right)}^2},$$ (3)

where p_i and o_i represent the individual predicted and observed values, respectively.

III. RESULTS

A. Effects of tone duration on detection performance in quiet

1. Psychometric threshold and DR

For each tone duration, the hit rate at each tone level and false alarm rate was used to compute the sensitivity (d′) as a function of tone level (see Sec. II E 1). The d′ vs tone level plots (psychometric functions) were used to compute the detection threshold (d′ = 1.5, after O'Connor et al., 1999) at that tone duration. The effect of tone duration can be seen in Fig. 1, which contains a large subset of the psychometric functions from the detection in quiet data set. The psychometric functions were shifted to lower tone levels for longer tone durations (Fig. 1), resulting in detection thresholds that decreased as the tone duration increased (Fig. 2). The effect of tone duration on detection threshold was confirmed by a mixed effects model that incorporated all of the detection thresholds in the detection in quiet data set (t = −7.6, df (degrees of freedom) = 257, p = 4.7 × 10⁻¹³). The model took the form threshold ∼ duration + frequency + frequency ∗ duration + (1|monkey). An interaction between the tone duration and frequency was not significant (p = 0.7). This suggests that whereas the tone frequency shifts the threshold up/down, the effects of the duration are not significantly different across the tone frequencies, which can be seen in the mean thresholds shown in Fig. 2.

FIG. 2.

(Color online) The psychometric thresholds (d′ = 1.5) and psychometric function DR (tone levels spanned by the dynamic portion of the function) averaged across monkeys as a function of tone duration at various frequencies [(A)–(D)]. The error bars represent one standard deviation.

It was also a goal of the present work to characterize the effect of tone duration on the psychometric function slope. In general, the slope increased as the tone duration increased, which was quantified using the DR or width of the psychometric function. The DR was inversely related to the slope (see Sec. II E) and was consistently less variable than the psychometric slope, likely due to the large fluctuations in d′ that occur at the ceiling performance in well-trained animals (Fig. 1). The effect of duration on DR/slope can be seen in the individual data in Fig. 1 and is summarized in Fig. 2. The DR decreased as tone duration increased. The effect of duration on the DR was confirmed using a mixed effects model that contained all of the data from the detection in quiet data set (t = −4.2, df = 257, p = 3.4 ∗ 10⁻⁵). In the mixed effects model, an interaction between the tone frequency and duration was not significant (p > 0.05), confirming that the effects of duration were not significantly different across the tone frequencies.

2. False alarm rates and RTs

In an effort to fully characterize the effects of duration on the performance, the RTs were measured as the time elapsing from the onset of the tone to the release of the lever on the signal trials (see Sec. II E 2). Many precautions were undertaken to ensure that the RTs were reliable (see Sec. IV A 3). The false alarm rates were inspected and analyzed as a function of the stimulus duration, and only the RTs to suprathreshold tones (35 dB SPL) were used to investigate the effects of the duration. The false alarm rates suggest very conservative decision criteria, making the likelihood of contamination of the RT data by guesses unlikely (see Sec. IV). Figure 3 shows the false alarm rates from all of the sessions used in these quiet and masked detection experiments (532 blocks of ∼200–265 trials per block). Figure 3 shows that false alarm rates were usually 0%–6% (over 65% of blocks), and the false alarm rates did not exceed 18%. Consistent with this, the stimulus duration did not affect the false alarm rate, suggesting that any effects of the stimulus duration on the performance (e.g., RTs) resulted from the stimulus, an issue which is elaborated on in Sec. IV. No effect of the duration on the false alarm rate was observed in a mixed effects model analysis, which included the false alarm rate as a dependent variable, tone duration and frequency as fixed effects, and individual monkeys as a random effect (effect of duration, t = 0.18, df = 516, p = 0.85).

FIG. 3.

The false alarm rates from all of the sessions (516 blocks of 200–265 trials, n > 100 000 trials) in quiet and noise. The total number of blocks for each monkey was 125 (monkey B), 112 (monkey G), 139 (monkey C), and 140 (monkey D).

In general, the RTs were longer for shorter tone durations when controlling for the effects of the tone level. Figure 4 shows the cumulative distributions of the RT for each duration. The RTs were to suprathreshold tones presented at 35 dB SPL across all of the tone frequencies tested. The effect of duration on the RTs was confirmed by constructing a mixed effects model that incorporated the tone frequency and duration as fixed effects and individual subjects as a random intercept term. The effect of duration on the RTs was significant (t = −6.3, df = 3697, p = 3.3 ∗ 10⁻¹⁰) as was the effect of the tone frequency (t = 4.36, df = 3697, p = 1.3 ∗ 10⁻¹⁰) but not the interaction between the duration and frequency (p = 0.76). To facilitate the comparison with the threshold by the duration functions that are typical in the literature, median RTs, pooled across monkeys, were plotted as a function of duration and are shown in Fig. 5.

FIG. 4.

(Color online) [(A)–(D)] The cumulative RT distributions for each monkey, pooled across the frequency. The tones were 35 dB SPL. The RTs to 3.25 (○), 6.5 (red ×), 12.5 (blue □), and 200 (green ⋄) ms are shown.

FIG. 5.

The median RTs to 35 dB SPL tones, pooled across monkeys, as a function of the tone duration. The RT by duration trend was consistent across frequencies, which are shown with different symbols (1 kHz, •; 8 kHz, ◻ and 32 kHz, ⋄).

B. Effects of tone duration on masked tone detection

1. Psychometric threshold and slope

The masked detection experiments were conducted to assess the effects of continuous, broadband noise (BBN), 76 dB SPL on temporal integration. The detection threshold decreased as the tone duration increased as with the detection in quiet data. The threshold by duration trends were remarkably similar in quiet and noise as illustrated in Fig. 6.

FIG. 6.

(Color online) The average psychometric thresholds (tone level at d′ = 1.5) from all monkeys normalized to the 200 ms threshold and plotted as a function of the tone duration. The quiet data are in black dashed traces, whereas the thresholds in noise are shown as solid green traces. The error bars represent one standard deviation.

The mean thresholds (averaged across monkeys), normalized to the threshold to 200 ms tones (usually the minimum threshold), are shown as a function of duration in Fig. 6 to illustrate how the performance in quiet compared to the performance in noise. This similarity held across the tone frequencies. The effect of the duration was confirmed with the mixed effects model analysis (p = 2.0 ∗ 10⁻³⁴), which contained an interaction term between the duration and frequency. As with the detection in quiet data, the interaction between duration and frequency was not significant (p > 0.05). The psychometric function slope increased as tone duration increased as with the detection in quiet data. This effect was quantified using the DR (the tone levels spanned by the dynamic portion of the psychometric function). The noise appeared to decrease the DR but only at shorter durations (3.25–12.5 ms). An example of the effect of masking noise on the DR is shown in Figs. 7(A) and 7(D). This effect of masking noise on the slope was confirmed with a mixed effects model containing the DR as the dependent variable, the tone frequency, duration, and noise as fixed effects, and individual monkeys as a random intercept term (effect of noise, t = 8.56, df = 514, p = 1.2 ∗ 10⁻¹⁶). The interaction between the background noise and duration in a mixed effects model was also significant and consistent with the observation that the noise only affected the DR at a subset of the tone durations (p = 0.005).

FIG. 7.

(Color online) The effect of the background noise on the DR. (A) The psychometric functions (d′ vs level) for detection of a 2 kHz tone of various durations in quiet by monkey B. (B) The exemplar psychometric functions when monkey B detected the 2 kHz tones of various durations in continuous 76 dB SPL noise. [(C)–(F)] The summary figures showing the mean DRs across all four monkeys as a function of the duration for 1 (C), 2 (D), 4 (E), and 8 kHz (F).

2. RTs

The RTs were also calculated for the masked detection performance. The RTs at 85 dB SPL (±5 dB, ∼20–35 dB above threshold) were separated by duration and monkey and medians calculated to provide an initial estimate of the speed in the task. Figure 8 displays the median RTs to 1 and 8 kHz tones pooled across the frequency, compared to the RTs in quiet (black traces). The RTs decreased with increasing duration, although qualitatively, the effect of the duration appeared weaker than suggested by the RTs to tones in quiet. The RTs were lower in noise, possibly due to the well characterized effects of sound level on RT (Kemp, 1984; Dylla et al., 2013). Specifically, the RTs decreased with an increasing tone level, even when the signal-to-noise ratio (SNR) is held constant. Thus, it is likely that the higher tone levels used in the masked detection experiments resulted in lower RTs. As with the detection in quiet data, the effect of the duration on the RTs was confirmed by constructing a mixed effects model that incorporated the tone frequency and duration as fixed effects and individual monkeys as a random intercept term. As with the RTs in quiet, the RTs during masked detection decreased with an increasing duration (t = −4.59, df = 3163, p = 4.6 ∗ 10⁻⁶) and increased with an increasing tone frequency (t = 10.3, df = 3163, p = 1.7 ∗ 10⁻²⁴).

FIG. 8.

(Color online) The median RTs as a function of the duration to 85 dB SPL tones in 76 dB, continuous, broadband noise (BBN; green traces), compared to the RTs to 35 dB tones in quiet (black traces). The open symbols show 1 kHz data, whereas the filled symbols represent 8 kHz data.

C. Phenomenological models

1. The power law vs exponential function

The thresholds from both the detection in quiet and masked detection data sets were fit with exponential and power law functions to provide estimates of the rate and range of temporal integration. We could find no systematic differences in the goodness of fit between the power law and exponential models or across the monkeys and tone frequency, which is similar to the results of O'Connor et al. (1999). This is illustrated by the scatterplot in Fig. 9, showing the BIC from each monkey at each frequency for the two models. The points generally cluster around the unity line, suggesting that the two models have approximately the same goodness of fit. Moreover, both models provided similarly good fits in terms of the root mean square error (∼1.5 dB, on average; Eq. (3) in Sec. II). For this reason, the exponential model was used for the estimation of different aspects of temporal integration to facilitate the comparison to previous macaque data, where the exponential function was found to be “most strongly descriptive of temporal integration” (O'Connor et al., 1999).

FIG. 9.

The BIC for the power law vs exponential functions for each monkey at each tone frequency to assess the goodness of fit of each power law or exponential function fit to the threshold data.

2. Estimates of temporal integration from the exponential function in quiet and noise

Three term exponential functions (see Sec. II) were fit to the threshold vs duration data from the data sets obtained for detection in quiet and BBN. An example can be seen in Fig. 10(A). The rate parameter τ, which is traditionally taken to be an estimate of the rate of integration, was extracted for each monkey at each frequency. Figures 10(B) and 10(D) show these values for detection in quiet and in noise, respectively. The constant of proportionality, I_k, provides an estimate of the range of thresholds from each exponential fit. These values are shown in Fig. 10(C). As it was a goal of this study to characterize the effects of noise on temporal integration, the exponential functions were fit to the masked detection threshold data as well, and τ and I_k were similarly extracted for the data obtained in the continuous background noise. The time constant and constant of proportionality values estimated during the masked detection can be seen in color in Figs. 10(D) and 10(E), overlaid on the estimates obtained from the detection in quiet data. Qualitatively, the estimates of the temporal integration rate (τ) and range (I_k) in quiet and noise look similar, suggesting that the noise does not have effects on these parameters. To validate this observation statistically, the effects of frequency and masking noise on the exponential model's estimates of the temporal integration rate were assessed by constructing a mixed effects model. The frequency and noise were entered as fixed effects, as was an interaction term between the two, to assess whether the effects of the tone frequency might be restricted to only one data set (quiet or noise). The time constants were not significantly different between the quiet and noise conditions (t = −1.17, df = 70, p = 0.25). Time constants were not significantly affected by tone frequency (t = −1.15, df = 70, p = 0.24). The interaction between frequency and noise was not significant, confirming that the effects of the frequency were not present in the quiet or noise data sets (t = 1.19, df = 70, p = 0.24). Similarly constructed mixed effects models confirmed that the constants of proportionality, which provide an estimate of the range of thresholds, were similarly unaffected by the frequency (t = −1.7, df = 70, p = 0.09) or masking noise (t = −0.45, df = 70, p = 0.65). The interaction term between the two was not significant, indicating that the lack of the effect of the frequency held true for both the quiet and noise data sets (t = −0.96, df = 70, p = 0.34). Figure 10(F) compares the time constants of the present study to the time constants estimated from the previously published data. The macaque time constants of the present study [open circles, Fig. 10(F)] were noticeably lower than chinchillas as reported by both Clark & Bohne (1986) and Wall et al. (1981) (green Δ), mice (×), cats (cyan stars), and previously published macaque (black line) time constants. The present data most closely resemble human (red ⋄) and budgerigar data (yellow □; see Sec. IV).

FIG. 10.

(Color online) (A) An example of the detection threshold decreasing with an increasing tone duration, fit with an exponential function. [(B),(D)]. The temporal integration rate estimates (the time constant, τ) from each exponential function are shown as a function of the tone frequency for all monkeys in quiet (B) and in noise (D). The symbols follow conventions in the legend in (C). [(C),(E)] The estimates of the range of thresholds from each exponential function are shown, based on the parameter, I_k. The parameters from the exponential fits to quiet data are in gray (C), and the parameters from the fits to masked detection data (E) are in color. (F) The temporal integration rate estimates (the time constant, τ) are compared to the previously published data from a range of species. The time constants to the previously published data were attained by curve-fitting with the exponential function [Eq. (2)] to thresholds in Fay (1988) for the uniformity of the methods and ease of comparison with O'Connor et al. (1999).

D. A probabilistic Poisson process model

Although the exponential and power law models are commonly used to describe temporal integration, they have limitations which have been described previously (Heil et al., 2017; Viemeister and Wakefield, 1991). In particular, they were not formulated based on the neurophysiological processes. This stands in contrast to process models such as the probabilistic Poisson model (PPM), formulated by Heil et al. (2017). Heil et al. (2017) described how the PPM could be used to model the performance in a yes-no task such as the one in the present report. Using their methods, the present data were also compared to simulated data generated using this model. Poisson probability density functions (PDFs) were generated using different rate parameters [Fig. 11(A)] to simulate the differences in the response evoked by stimuli of different intensity/duration combinations. The detection theoretic methods (receiver operating characteristic, ROC, analysis) were then used to model the detection performance that could be based on such responses. Namely, placement of a decision criterion at a given event rate and calculation of the area under the signal distributions to the right of the decision criterion (red line labeled c) yielded hit rates (Fig. 11). The area under a noise distribution [Fig. 11(A)] to the right of the decision criterion yielded false alarm rates. The decision criterion (ten events per second) and rate parameter of the noise distribution (five events per second) were selected to produce a false alarm rate that matched our typical false alarm rates of ∼5%. This parameter choice is very similar to that in Heil et al. (2017), who used very low-rate noise distributions that matched the low-spontaneous rate auditory nerve fibers (ANFs) to provide the best match to the behavior. The choice of low event rates for the noise distributions poses an interesting theoretical issue. Namely, it is widely held that high-spontaneous rate fibers not low-spontaneous rate fibers support the detection of quiet signals (e.g., Costalupes, 1985). Model psychometric performance similar to what we report could still be achieved with higher event rate noise distributions that would mimic a combination of low- and high-spontaneous rate ANFs. Use of such a noise distribution simply required an increase in the rate of the signal distributions (suggestive of the integration of larger numbers of ANFs), but the same psychometric threshold and slope trends that held with low event rate noise distributions held for this case also (data not shown).

FIG. 11.

(Color online) (A) The Poisson PDFs with rate parameters corresponding to the noise (no stimulus) distribution and responses to 200 and 50 ms stimuli. “c” is the decision criterion used in the standard detection theoretic analyses to calculate hits and false alarms. (B) The receiver operating characteristic (ROC) curves generated by integrating the area under Poisson PDFs at different decision criteria (“c”) are shown.

The event rate parameter of the signal distributions [Fig. 11(A)] were selected to most closely evoke the simulated performance, similar to what we observed in our behavioral experiments, at a range of the intensity/duration combinations. This was performed by manipulating the rate parameter as a means of generating more simulated events or spikes. For longer durations (100 and 200 ms), the simulated behavioral performance could be generated using spike counts of about 200, which, assuming a maximum firing rate of 350 spikes per second, suggests integration of 3–5 ANFs (see Sec. IV). For shorter durations, greater firing rates were used to increase the number of events to 75 for 50 ms stimuli, ∼60 for 25 ms stimuli, and about 20 for 12.5, 6.5, and 3.25 ms stimuli. For 3.25 ms stimuli, given a maximum firing rate of about 350 spikes per second for each ANF, the ceiling psychometric performance suggests the integration of ∼20–25 ANF firing rate distributions (see Sec. IV). To facilitate the comparison with behavior, such as shown in Fig. 1, it was desirable to display the model performance as a function of the tone level instead of the firing rate or spike count. To accomplish this, we related the event rate to the tone level with the rate-level model of Sachs and Abbas (1974). This allowed the generation of psychometric functions [Fig. 12(B)] that showed the same threshold and slope trends as our empirical psychometric functions [example in Fig. 12(A)]. In this specific case, the psychometric functions are displayed in terms of the hit rate rather than d′ to facilitate the visual comparison with the model data in the subsequent panels in Fig. 12. The hit rate in Fig. 12(B) corresponds to the area under the Poisson distribution. A more typical d′ measure was not used because d′ assumes an underlying normal distribution, which does not apply to this model. The threshold was taken to be the tone level that evoked a 0.5 hit rate (area under the curve), a common threshold criterion when extracting the threshold from hit rate functions (Beitel et al., 2003; O'Connor et al., 1999). The psychometric thresholds (in terms of d′) for all of the monkeys detecting 2 kHz tones in the quiet and model thresholds (hit rate = 0.5) are compared in Fig. 12(C). A related question is whether the PPM could be adapted to mimic the behavior in noise with the main finding in the present report being that noise decreases the psychometric DR. To some degree, this is predicted by the model. By shifting the distributions to higher baseline rates to simulate the presence of noise, the DRs contracted for short duration stimuli as is similar to what was observed behaviorally. However, model thresholds in noise did not match behavioral thresholds in noise, at least when a necessary, realistic parameter of noise-induced changes in nerve fiber responses was imposed upon the model. Specifically, Costalupes et al. (1984) and Gibson et al. (1985) documented that background noise caused a 0.61–0.79 dB/dB threshold shift in that ANFs, which does not match the 1 dB/dB behavioral threshold shift that macaques and humans exhibit (Dylla et al., 2013; Hawkins and Stevens, 1950). Thus, the model captures one aspect of the behavioral data in the background noise.

FIG. 12.

(Color online) (A) The example psychometric functions from monkey B at 2 kHz (x axis range limited to facilitate the visual comparison with the model data; original axes are shown as an inset). The hit rate, used to calculate d′ in previous figures, is displayed to facilitate the comparison with the model data (see Sec. III D). (B) The psychometric functions generated using the PPM outlined in Fig. 8. The threshold is the tone level that evoked the hit rate = 0.5. (C) Comparing model thresholds to behavioral thresholds. Model thresholds are displayed as black diamonds connected by red lines, whereas the behavioral thresholds from all of the monkeys at 2 kHz are displayed as gray squares.

The PPM consistently yielded slightly worse fits than the exponential and power law models, which is indicated by the greater root mean squared error (RMSE) values [Eq. (3) in Sec. II], which were consistently around 1.0–1.5 dB for the exponential and power law models, whereas the PPM fits were more consistently 2–3 dB for the PPM (Table I). Nonetheless, the PPM gives insight into a potential biological substrate of temporal integration, namely, that event rates suggestive of the integration of multiple ANF responses is required to match the behavioral performance, especially for short-tone durations. This insight can guide future neurophysiological studies in macaques.

TABLE I.

The final parameters for fitting the PPM to the psychometric data. The key parameter for fitting the model to various sets of data (the range of the rate-level function used for sound level conversion; see Sec. III D) is indicated for thresholds at a range of different tone frequencies. The range parameters that gave the lowest average RMSE is indicated, along with the average RMSE for all four monkeys. The RMSE for the exponential fits were consistently 1–1.5 dB (no effect of the tone frequency).

Tone frequency (kHz)	Tone level range (dB)	Poisson model RMSE (dB)	Exponential model RMSE (dB)
1	0–45	3.19	1.28
2	0–25	2.9
4	−5–25	2.98
8	−5–20	2.39
16	−5–15	2.59
32	10–60	1.89

IV. DISCUSSION

A. Comparison with other psychophysical studies

1. Threshold

Consistent with other studies of the effect of signal duration on the detection across a variety of species, we found that threshold decreased exponentially as tone duration increased (e.g., mice, Ehret, 1976; porpoise, Johnson, 1968; macaques, Clack, 1966; O'Connor et al., 1999; humans, Plomp and Bouman, 1959; Watson and Gengel, 1969; budgerigars, Wong et al., 2019; blue monkeys and mangabeys, Brown and Maloney, 1986). Although the threshold data in the current report are in agreement with many aspects of these studies, the τ values (temporal integration time constants) in the current report were 10–40 ms, in contrast to the macaque data of Clack (1966) and O'Connor et al. (1999), who found τ values in the range of 140–150 ms. Those earlier data suggest that macaques exhibit drastically slower temporal integration than humans (human mean of ∼30 ms; Watson and Gengel, 1969; Clack, 1966; reviewed in O'Connor et al., 1999; O'Connor and Sutter, 2003; Recanzone and Sutter, 2011). The present data suggest that the macaque data resemble the data from humans and budgerigars [Fig. 10(F)]. The evaluation of this claim is critical to determining the extent to which macaques can serve as a model of human auditory perception because of the potentially specialized nature of human temporal processing (Gai et al., 2007; O'Connor et al., 1999;O'Connor et al., 2011). The present data suggest that macaque and human temporal processing, as assessed by this metric, are very similar. This may relate to macaque and human performance in modulation frequency discrimination tasks, which suggest primates have an advantage over rodents (Fig. 9 in Moody, 1994). But, it is curious that other previous studies have not found this advantage (O'Connor et al., 1999; O'Connor et al., 2011). A strong candidate explanation for the difference in the current report and the two earlier macaque studies may lie in the range of tone durations at which the detection was measured. The current study used 3.25–200 ms tones, whereas O'Connor et al. (1999) used 25–800 ms tones. This may have caused the monkeys in the present study to integrate at faster rates resulting from their being exposed to and trained on shorter duration stimuli. Moreover, a significant consideration is whether the spectral splatter of the shorter duration tones (3–6 ms tones had 1.25 and 2.5 ms ramps, respectively) contributes to the patterns of behavioral performance reported here. However, this is unlikely to explain the differences in the threshold trends as the temporal integration time constants reported here were virtually identical to those from analyses that used only data from tones 25 ms and longer (data not shown).

It was also a goal of this study to characterize the effects of tone frequency on temporal integration, as the literature is relatively unclear in this respect. Some studies have reported a lack of effect of tone frequency (Ehret, 1976; O'Connor et al., 1999), while others have found this relationship to follow an inverted U-shape (Clack, 1966; Johnson, 1968). The present data suggest no consistent effects of tone frequency on threshold measures of temporal integration in quiet or noise (Sec. III C 2). Mixed effects model analysis of thresholds, rather than temporal integration time constants, corroborated this finding with a lack of interaction between tone frequency and duration.

The exponential function's range parameter, I_k, was also a focus of this study. The mean I_k value for detection in quiet [Fig. 7(C)] was very comparable to the mean of ∼20 dB reported by O'Connor et al. (1999) for macaques. The data from cats are less similar in this respect (<10 dB), at least at the very highest frequencies (Costalupes, 1983), and may reflect a species difference. However, at lower frequencies, the magnitude of threshold change appeared to be larger (∼15 dB) and within the range of the macaque data. The range of threshold values observed in humans appear to be more similar to those in the current report (∼20 dB; Fig. 9(D) in O'Connor et al., 1999).

2. Psychometric function DR/slope

The psychometric function DR changed as a function of tone duration, indicating that the variability of the performance increases at shorter stimulus durations. This is consistent with human reports of DR (or its inverse, the psychometric function slope) indexing the performance variability (Chauhan et al., 1993; Dahmen et al., 2010) and with previous reports from this laboratory. Namely, DR is sensitive to statistical properties of the masking noise (Rocchi and Ramachandran, 2018). The current report also investigated whether the background noise modified the DR changes as a function of tone duration. Whereas there were no discernible effects on the rate or range of threshold measures of temporal integration, there were strong, consistent effects of noise masking on the DR. The DR decreased, which is consistent with neurophysiological studies that have reported compression of the DR of the ANFs in the presence of continuous noise (Costalupes et al., 1984; May and Sachs, 1992; Palmer and Evans, 1982). Although the PPM (Sec. III D) could mimic this effect, it could not reproduce the threshold shift due to the noise when a necessary, realistic parameter was imposed on the model (the 0.6–0.79 dB/dB threshold shift documented by Costalupes et al., 1984, and Gibson et al., 1985). The behavioral threshold shifts in macaques and humans are 1 dB/dB at and above 1 kHz (Dylla et al., 2013; Hawkins and Stevens, 1950), implying that transformations in the central auditory pathway are necessary for detection in noise.

The effect of noise on the psychometric DR speaks to the importance of using measures in addition to threshold when attempting to elucidate the effects of noise on auditory processing. Considering the psychometric function slope or DR may become increasingly important in understanding the effects of hearing impairment as recent hidden hearing loss studies have demonstrated a lack of the effect on threshold measures of temporal integration (Marmel et al., 2020; Wong et al., 2019) but did not report the psychometric function slopes/DR. The PPM presented here, along with previous models (Lopez-Poveda, 2014; Stevens and Wickesberg, 1999), suggest that reduced auditory nerve input could have such perceptual effects and could be a fruitful topic for future studies.

3. RT

Consistent with human psychophysical studies in the visual system (e.g., Hildreth, 1973; Kietzman and Gillam, 1972; Miller and Ulrich, 2003), somatosensory system (Hernández-Pérez et al., 2020), and a single study in the auditory system (Raab, 1962), we found that RT increased as the stimulus duration decreased. This extends the currently available data in animal models to show that insofar as these behavioral paradigms index the temporal integration, the process also occurred at suprathreshold levels. To the best of the authors' knowledge, this has not been shown before in any animal model of auditory perception. The data from chinchillas and cats demonstrate a lack of effect of sound duration on RT and, thus, the RTs serve as the second line of evidence suggesting human-macaque perceptual similarity (chinchillas, Clark, 1979; cats, Costalupes, 1983). The data in the current report more closely resemble the data from humans (Raab, 1962; compare to Fig. 4). Although the threshold measures of temporal integration did not exhibit effects of frequency in the present data, RTs were strongly affected by frequency (Secs. III A 2 and III B 2). This is consistent with the sparse human and macaque literature reporting effects of tone frequency on RTs (Chocholle, 1940; Pfingst et al., 1975).

The use of RTs in yes/no detection tasks is problematic for reasons others have detailed (Tiefenau et al., 2006). Namely, some responses on signal trials may be uncaused by the stimulus because it is near the detection threshold, making these responses a kind of “false alarm.” These false alarms can be referred to as “signal trial false alarms” to differentiate them from the false alarms that are calculated from the rate of responses on the catch trials. Tiefenau et al. (2006) constructed a race model to account for the signal trial false alarms and the contaminating effect they have on the RT distributions. To account for this problem in the present analysis, only the RTs at ∼35 dB SPL (∼20–35 dB above threshold) were included. This increases the likelihood that hits were caused by the stimulus rather than a false alarm process because the tone level is substantially suprathreshold. All of this suggests that the RT distributions constructed from the hit trials were not significantly contaminated by signal trial false alarms as might be the case in other data sets where the false alarm rates are higher. Moreover, if the effects on RT observed here were due to signal trial false alarms described by Tiefenau et al. (2006), one would expect tone duration to have an effect on the false alarm rate. No such effect was found (see Sec. III A 2).

The effects of duration on RT shown here are presented with the caveat that there is an interaction of the level and SNR on the RTs in the detection experiments (Kemp, 1984). This is relevant to this analysis because the duration affects threshold and, thus, the SNR. A simple solution would be to hold the SNR (dB re: threshold) constant and change signal duration; however, when Kemp (1984) held the SNR constant, the effects of absolute tone level remained. It is not possible, then, to empirically manipulate the duration without the accompanying effects of level or SNR. For this analysis, level was held constant at the expense of SNR varying as the tone duration did, and this could arguably complicate the interpretation of the data. Given this caveat, the effects of duration on RT are consistent with theoretical predictions of temporal integration and reports from groups studying the auditory, visual, and somatosensory systems.

B. Potential neurophysiological mechanisms

How temporal integration is implemented in the nervous system has been the subject of great discussion over the past two decades. Although the current results do not provide direct evidence for or against a particular theory on their own, the modeling results may help guide future neurophysiological studies of the temporal integration. The close match between the PPM data and psychometric data presented here is consistent with the interpretation that a probabilistic decision process may underly the detection behavior as others have suggested (Heil et al., 2017; Meddis and Lecluyse, 2011; Viemeister and Wakefield, 1991). More specifically, it suggests that integration of ∼20–25 ANF firing rate distributions is necessary to account for the behavioral performance at the shortest duration presented here given that ANF firing rates can reach 250–350 spikes per second (Heil and Peterson, 2015; Yates et al., 1990; Nomoto et al., 1964). When previous studies have experimentally investigated the neuronal correlates of auditory temporal integration, conclusions have widely differed across studies. A study of single-units in the cochlear nucleus of chinchillas found a similarity between the behavioral integration times and those estimated from firing rate-based thresholds (Clock et al., 1993). The data from a follow-up study in the auditory nerve found the integration times to be much slower than those estimated from the behavior (Clock Eddins et al., 1998), suggesting a transformation at the level of the cochlear nucleus, which is consistent with the model presented here requiring integration of multiple ANFs, anatomical evidence of which exists for certain neuronal populations in the ventral cochlear nucleus of cats (e.g., Smith and Rhode, 1987). Other models constructed to account for the perception of short duration stimuli have included this same kind of across-fiber integration (Lopez-Poveda, 2014; Stevens and Wickesberg, 1999) This suggestion is made with the caveat that species differences in the organization of other regions of the cochlear nuclei exist between cats and primates (Moore, 1980, but see Rubio et al., 2008). An alternative theory based on the single-unit data in cats suggests that first spike latencies in the auditory nerve may be the neurophysiological basis for detection and, thus, temporal integration (Heil et al., 2008). Different still, the evoked responses in humans have been used to argue that the neural correlates of the temporal integration are beyond the primary auditory cortex (Lütkenhöner, 2011). Future studies could expand on these findings by recording from animals engaged in a detection task as both anesthetic agents and task engagement affect the neuronal encoding of sound (Downer et al., 2015; Ramachandran et al., 1999; Rocchi and Ramachandran, 2020; Ryan and Miller, 1977; Slee and David, 2015).

V. SUMMARY AND CONCLUSIONS

The tone detection thresholds, RTs, and DR of the psychometric function estimated from macaque behavior all converge to provide evidence of temporal integration. The time constant of temporal integration suggests that human and macaque integration rates are very similar, in contrast to previous reports. This similarity could prove crucial in understanding how changes in this process following noise exposure contribute to speech processing deficits. The first effect of sound duration on the RT in an animal model is also reported here, which will provide a basis for further investigation of how different stimulus attributes probe the speed of auditory processing as measured at behavioral and neuronal levels. These data also speak to the effects of noise on these three psychophysical measures. Whereas the threshold and RT measures of temporal integration did not appear to be sensitive to the effects of noise, the psychometric DR did appear to be sensitive. This suggests that the DR may be a more sensitive measure of hearing in noise than other measures and important in future studies of processing deficits in noise caused by aging, ototoxicity, and noise exposure. More generally, these data suggest that macaques may be uniquely qualified to model human temporal integration because, in this case, unlike previous reports in animal models, multiple psychophysical measures converge to be consistent with the predictions of models of temporal integration. This comprehensive psychophysical approach in macaques could prove useful in future studies of hearing loss intended to further inform existing diagnostic and therapeutic strategies, and the modeling results point the way forward for future neurophysiological studies.