Tonotopic Selectivity in Cats and Humans: Electrophysiology and Psychophysics

For the psychophysics, we kept the frequency spectrum and stimulus level of the maskers the same as for the COR recordings, and presented them continuously throughout each threshold estimation. The probe level was then adjusted in order to derive probe detection thresholds for the different masker types (1/8th-octave, 1-octave and notched-noise masker). Detection thresholds with the notched-noise masker were also measured for both cats and humans.

Four male cats were trained and yielded psychophysical data; they were neutered to reduce aggressive behaviour and thereby enable group housing. Cats 1–3 yielded complete data in all conditions. Cat 4 was tested with all the masker conditions but only with the 8-kHz probe tone. The cats were 5–16 months of age at the beginning of training. Weights ranged from 4.1 to 5.9 kg during most of the data collection. Food was restricted on psychophysical training days (5 days/week). On those days, cats received all or most of their food as moist food provided as reward during the psychophysical task. They generally worked until sated, but dry food and water was offered freely for ~ 1 h after each testing session. On weekends, cats received free access to dry food for 3 h per day. Water was freely available in the housing area.

The cat psychophysical experiments were conducted in a double-wall sound-attenuating chamber, having interior dimensions of 2.6 × 2.6 × 2.5 m, and lined with SONEXone adsorbent foam to suppress sound reflections. The cat sat or stood on an elevated platform in the centre of the chamber. A harness restrained the cat to the platform while permitting free movement of the head and limbs. Cats generally maintained an orientation of the head and pinnae toward the sound source, located in the front of the chamber. A pedal was positioned in front of the cat for psychophysical responses, and a feeder was present that could deliver small portions of canned commercial cat food. Acoustic stimuli were presented through a 3″ co-axial loudspeaker (Fostex FF85WK) in a bass reflex enclosure located 1.2 m in front of the cat. Similar to the COR recordings, stimulus generation, data acquisition, and experimental control used TDT equipment controlled by custom MATLAB scripts on a Windows-based personal computer. The speaker was calibrated daily in the same way as for the COR recordings. The detection task used a hold-release design that was derived from that described by May and colleagues (1995) and adapted in our previous study (Javier et al. 2016). Each trial began by the operator illuminating a green light near the location of the sound source. The green light signalled to the cat that it could press and hold a pedal to activate a continuous sound consisting of just the pink noise (on no-mask trials) or the pink noise plus the 1/8th- or 1-oct masker. After a delay varying randomly among values of 2, 3, 4 and 5 s, the probe sound was initiated, lasting to the end of the trial; the various probe-onset conditions are referred to, respectively, as Hold 2, Hold 3, Hold 4 and Hold 5. A release of the pedal earlier than 1000 ms prior to probe onset was recorded as an “early release” and was not scored. A release 1000 to 0 ms prior to probe onset was scored as a “false alarm”. A release of the pedal within 0 to 1000 ms after probe onset was scored as a “hit”, and the cat received a reward consisting of a small portion of food. Releases later than 1000 ms after probe onset were scored as “misses”. Early releases, false alarms, and misses were punished by a 2-s time-out period signalled by a flashing blue light in which no trial could be initiated. The masker was terminated after each pedal release, or 1000 ms after the end of the hit window, to be restarted by the next pedal press.

Figure 6 shows the latencies of pedal-release times for one cat in the various Hold conditions; data here are combined across all masker conditions and probe frequencies and levels, and colours indicate how the various latencies were scored. The histograms in the figure include trials both in which the probe level was above threshold and in which the probe level was below threshold, which accounts in part for the large number of misses. Note that the false-alarm response-time window for Hold N + 1 trials coincided exactly with the hit window of Hold N trials. For that reason, we used the false alarms on Hold N + 1 trials as catch trials (i.e. no probe) for the Hold N trials. Hold 5 trials were used only as catch trials for Hold 4; hits on Hold 5 trials were rewarded but were not included in the computation of hits. Hit rates were counted separately for each masker type, probe frequency, and probe level condition, whereas false alarm rates were counted separately for each masker type but combined across all probe frequencies and levels. The rationale for combining false alarms across probe conditions is that the cat produced a false alarm or correct rejection prior to the onset of the probe, rendering any probe properties irrelevant for the count of false alarms. The proportions of false alarms and hits relative to misses tended to increase with increasing hold time, which we attribute to cats’ impatience to receive the food reward. We aimed to mitigate any bias that might have been introduced by those trends. For that purpose, we computed hit and false-alarm proportions rates separately for each hold time; those numerical values are given in the illustration. For each block of trials, a false-alarm rate was computed separately for Holds 3, 4 and 5, those rates were weighted according to distribution of Holds 2, 3 and 4 tested for each probe condition, and then a weighted average false alarm rate was computed.

The sound levels of the 1/8-oct and 1-oct maskers were 65 dB SPL and 73 dB SPL, respectively. Those levels were chosen to equalise the energy within the human-sized auditory filter; the human equivalent rectangular bandwidth (ERB) centred on 8 kHz was estimated to be 888 Hz (Glasberg and Moore 1990). In the present psychophysical experiments, probe levels on each trial were controlled by the operator and were adjusted adaptively to fill out psychometric functions spanning sub-to-suprathreshold levels in 2-dB increments. Test sessions were carried on across days until enough trials were completed at each masker, probe, and level condition, with a minimum of 2 probe levels tested above and 2 below the final threshold. A minimum of 15 trials was obtained at probe levels that were well below or above the eventual measured threshold and a larger number was obtained at levels within 2 dB of that threshold. Typically, the probe frequency and masker condition were held constant throughout a session, with only the probe level varying among trials.

Performance was given by the unbiased maximum proportion correct, P(c)max (Green and Swets 1966; Macmillan and Creelman 1991). For each stimulus condition:where P_hit and P_{false alarm} are the proportions of hits and false alarms, z is the transform to standard deviates, d’ is the unbiased sensitivity index, and Φ is the normal distribution function. Logistic equations were fit to plots of P(c)max versus signal level, and the thresholds for detection of tones were given by the interpolated signal level at which the logistic curve crossed P(c)max = 0.69; that value corresponds to d’ = 1. We estimated the chance level for P(c)max by performing a permutation test in which we randomised the association between Hold times and pedal releases. That test showed the 95th percentile of P(c)_max given chance performance as 0.61.

Group B of the human participants (minus author AJH, N = 11) undertook the psychophysical procedures and completed all the psychophysical trials. Experiments were conducted with the listener seated in a sound-attenuating booth. In a similar manner to the COR recordings, all stimuli were generated using custom MATLAB scripts, played through an RME Fireface UCX sound card, TDT headphone drivers, and a monaural ER2 insert phone. The particular ear that was tested was balanced among listeners (6 left, 5 right). Stimuli were calibrated using an ear coupler (GRASS) and a Hewlett Packard 303 HP3561A dynamic signal analyser.

Masker levels were held at 60 dB SPL for the 1/8th-oct masker and 67.9 dB SPL for the 1-oct and notched-noise masker; those masker levels were equal to those used in the COR recordings. A 2-alternative forced choice adaptive procedure was used in which the probe level was increased after every incorrect response and decreased after every two consecutive correct responses. The change from increasing to decreasing probe level or vice versa was defined as a turnpoint. Each run of the procedure finished after 8 turnpoints, with the threshold for that run being computed from the mean of the last 6 turnpoints. The step size was 6 dB for the first 2 turnpoints, and 2 dB thereafter. The 2- down 1-up adaptive procedure estimates the signal level yielding 70.7 % correct (Levitt 1971). A noise stimulus consisting of the pink noise with or without the 1/8th-oct, 1-oct or notched-noise masker was presented continuously throughout each adaptive track. The probes were 50-ms long with 2-ms sine-squared ramps, as in the COR recordings. In each trial, two windows flashed successively on a computer screen, and the probe was presented randomly during one or the other window. The listener’s task was to report the time window that coincided with the probe. Feedback was given to the listener after each trial. Thresholds were computed from the average of 2 runs, or 3 if there was more than a 5-dB difference between the first two thresholds.

We are aware of three types of method that have been used to measure selectivity with scalp potentials, all of which were developed in experiments with human participants. One paradigm measures the response to a tone of a given frequency in the presence and absence of intervening tones at varying frequency separations (Butler 1968, 1972; Picton et al. 1978; Näätänen et al. 1988; de Boer and Krumbholz 2018). This has led to mixed results. Some authors reported very broad tuning (Butler 1968, 1972; Picton et al. 1978). For example, Butler (1972) found that the COR to 250-Hz tones presented every 5 s could be reduced by intervening 8-kHz tones. One feature of this method is that there is a (usually long) silent gap between the test and intervening tones, so that there is no masking per se between the intervening tones and test tones at the level of the cochlea or auditory nerve. Rather, the method relies on adaptation at multiple, more-central, stages of the auditory system, including at the cortex. Differences in stimulus parameters may cause the onset response to be dominated by different classes of neurons, which may in turn differ in their tonotopic selectivity. Indeed, the study by Näätänen et al. (1988) used a much shorter SOA of 500 ms and reported considerably sharper tuning than studies that used longer SOAs (e.g. Butler 1972). The measures used here involve masking of the neural response to the probe at the level of the cochlea and auditory nerve, and used a fairly short SOA of 1 s.

A second method is to change the frequency of a tone part-way through a stimulus, and to measure the size of the resulting COR as a function of the frequency separation between the two tones. He et al. (2012) measured this “acoustic change complex (ACC)” in 12 normal-hearing adults in response to changes in the frequency of a 500-Hz tone. The mean threshold frequency change was 5.8 Hz, equal to 1.4 % of the baseline frequency. Although larger than the psychophysical thresholds obtained in the same subjects, this value is much smaller than the smallest frequency separation of 19 % (i.e. 1/4th oct) between the probe and the centre frequency of the 1/8th-oct masker in our experiments. The ACC is the only scalp-recorded measure of cortical selectivity that we are aware of that has also been applied to a nonhuman species: Presacco and Middlebrooks (2018) obtained pure-tone ACC thresholds of between about 1 and 5 % in the sedated cat and showed that these thresholds were similar to psychophysical measures reported previously for the same species. The ACC can also be elicited in CI users by changing the stimulating electrode mid-way through a stimulus (e.g. He et al. 2014; Mathew et al. 2016). An advantage of the ACC measure over our masking method is that, because only one stimulus is present at any one time, it avoids complications associated with masker-probe interactions such as beating and cochlear distortion products for acoustic stimulation (Moore et al. 1984) and charge interactions for electrical stimulation (Cosentino et al. 2015). A disadvantage is that an ACC can be generated to changes in intensity and/or loudness, making it important to loudness-balance the stimuli. Loudness levels can vary substantially from electrode to electrode in CIs, especially for focussed modes of stimulation (Bierer and Faulkner 2010), making the procedure time-consuming in humans and hard to achieve in animal experiments. As we wished to generalise our method to CI experiments with focussed stimuli, this was one reason for not choosing the ACC paradigm.

Third, some authors have measured electroencephalography (EEG) or magnetoencephalography (MEG) responses to the onset of tones presented in noise containing different notch widths, analogous to our COR measures with notched-noise maskers (Sams and Salmelin 1994; Kauramäki et al. 2007; Okamoto et al. 2007). Those studies were primarily interested in the effects of attention, but included one condition similar to ours in which the subjects listened passively. Kauramäki et al. (2007) used EEG and measured the global field power of the N100 response to a 1-kHz tone presented in quiet or in noise containing notches of different widths centred on 1 kHz. Their COR increased with increases in Δf up to 25 % at which point it was similar to that in the no-mask condition. This differs from our notched-noise experiment in which a substantial (6 dB) reduction in COR was observed even at the Δf of 30 %. In addition to using lower noise and probe levels than here, Kauramäki et al. increased notch width by progressively removing portions of the masking-noise spectrum, such that its total bandwidth and overall power decreased with increasing notch width. The same was true of the study by Sams and Salmelin (1994), who measured the MEG N100m response to the onset of 1- and 2-kHz tones in notched-noise backgrounds. They observed quite variable and sometimes nonmonotonic functions relating the N100m to notch width. Our protocol differed from that of the preceding two studies in that we maintained a constant width of the upper and lower masker bands as they were separated to produce progressively wider notches, consistent with the stimuli used in psychophysical notch-noise-masking experiments described here and elsewhere. Okamato et al. (2007) found that the N1m to amplitude-modulated 1-kHz tones masked by notched noise increased up to the maximum notch width studied, but this corresponded to only an 8 % distance between notch edge and probe frequency, so we do not know how the N1m would have varied over wider ranges of notch widths.

A stimulus consisting of two primary frequency components f1 and f2 can produce cochlear distortion products (DPs), the largest of which are typically the quadratic and cubic DPs having frequencies of f2-f1 and 2f1-f2, respectively, and that can sometimes be perceived by listeners (Goldstein 1967; Hall 1972; Smoorenburg 1972a; Zwicker 1979, 1981). Although they are usually studied using pure tones, DPs can also arise from the combination of a pure tone and a narrow band of noise, such as the 1/8th-octave maskers used here. Our pink noise encompassed the predicted frequencies of the quadratic DP for all masker and probe combinations, but, for the 1/8th-octave masker this was true for the cubic DP (CDP) only when the signal frequency was 0.5 octaves below the masker centre frequency.

The amplitude of the CDP is generally largest with small f2/f1 ratios and (for human listeners) becomes inaudible at ratios of between about 1.4 and 1.6 (Smoorenburg 1972b). We therefore think it unlikely that the CDP would have contributed to the COR observed for masker-probe separations of plus or minus 0.5 octaves, corresponding to a ratio of 1.41; furthermore, when the probe was 0.5 octaves above the signal the CDP would have been masked by the pink noise. However, the ± ¼-octave separation used for two probe frequencies corresponds to a ratio of 1.18 at which the CDP can be large. For our human participants, the centre frequency of the CDP for 3364-Hz probe would have been 2728 Hz, which is more distant than the 3364-Hz signal from the 4000-Hz masker, and so could have in principle contributed to the COR. We conclude that the CDP may have contributed to the COR for signals that were ¼ octave below the masker centre frequency but was unlikely to have done so for other signal frequencies. In the cat, we performed a control experiment with one additional cat with the pink noise cutoff extended to 7190 Hz (i.e. above the expected CDPs at 5454 and 6486 Hz, because of the doubling of the frequencies in the cat). That noise reduced the response to the 8000-Hz probe in the absence of any other masker, so we reduced its level by 10 dB, added it to the 1/8th-oct masker, and measured the COR to each masker/frequency combination. This showed a pattern of COR with probe frequency that was typical of those observed in the main experiment.

The psychophysical masking experiments with cats found that the 1/8-oct band at 65 dB SPL produced about the same amount of masking at 8 kHz as did the 1-oct noise band, which had a level of 70 dB SPL for five cats and 68 dB SPL for two cats. When we converted these thresholds to signal-to-masker ratios (SMRs) at the output of the cat auditory filter, using the estimates of filter ERB obtained from the notched-noise measurements, we found that the SMR was on average 1.9 dB higher for the 1-oct than for the 1/8-oct masker. This “excess masking” was statistically significant (t(3) = 3.96, p = 0.029). A similar analysis for the human listeners (and using the human ERB) found no significant difference. However, a two-way ANOVA on the masked thresholds for the probes at the masker centre frequency, relative to the masker level per species-specific ERB did not find a significant effect of species or a species X masker interaction. Hence, although we found significant excess masking only for the cat, the absence of a significant interaction means that we have no evidence that the excess masking differed significantly between species.

Our observation that the feline auditory filter is approximately 22 % wider than that of humans is consistent with several previous reports. Using a band-widening procedure with cats, Pickles (1975) and Nienhuys and Clark (1979) estimated critical bands at 8 kHz as around 1520 or 1290 Hz, respectively. Those values are in line with our value of 1087 Hz, considering that critical-band estimates tend to be broader than ERBs obtained with notched noises. In another member of the Carnivore order, the ferret, a notched-noise study yielded an even wider ERB at 8 kHz of 1680 Hz (interpolated from values for 7 and 10 kHz reported by Alves-Pinto et al. 2016). We note that our study employed simultaneous masking, which tends to yield broader ERBs than do forward-masking procedures (Oxenham and Shera 2003; Sumner et al. 2018); however, simultaneous masking was also used in the other animal studies cited here (Pickles 1975; Nienhuys and Clark 1979; Alves-Pinto et al. 2016) as well as all the human studies considered in the derivation of an equation for ERB versus centre frequency by Glasberg and Moore (1990). Measurements of ERB using notched noise centred at 8 kHz in species less closely related to cats have yielded ERB values ranging from ~ 900 Hz (Evans et al. 1992), through 1377 Hz in macaque (Burton et al. 2018), to 1540 Hz in mouse (May et al. 2006).