Frequency Selectivity in The Parakeet Studied by Isointensity Masking Contours

There are a number of masking procedures currently in use for behaviourally estimating the frequency selective properties of the auditory system (Zwicker, 1974; Vogten, 1978a, b; Moore, 1978). These procedures are important because frequency selectivity, which describes the capacity of an organism to discriminate frequency components within a complex sound, is one of the fundamental processes of hearing. Furthermore, the behavioural measures of frequency selectivity can be compared with filter processes observed in the cochlear receptor or auditory nerve to ascertain the central or peripheral contributions to frequency selectivity. While there is considerable knowledge of frequency resolving processes in mammals, relatively little is known of this aspect in non-mammals, and this is especially true of birds. Nevertheless, the masking procedure used by Wegel & Lane (1924), narrow and broad band noise masking, and psychophysical tuning curves have been used to examine frequency selectivity in the parakeet (Saunders, 1976; Saunders, Bock & Fahrback, 1978a; Saunders, Else & Bock, 1978c; Saunders, Denny & Bock, 1978b; Saunders, Rintelmann & Bock, 1979a). These data represent a substantial portion of the information on frequency resolution in birds. There is yet another method in this masking arsenal that has been used less frequently, but which has a number of important features. This procedure provides data that represent isointensity masking contours (Vogten, 1974), and these are valuable since they are homologues of response area curves measured from single fibres of cells in the auditory pathway.

The present study was designed to examine isointensity masking contours in the parakeet in order to see if this measure of frequency selectivity matched the unique aspects of frequency selectivity revealed by other masking procedures in this species.

Four parakeets (Melopsittacus undulatus), aged between 6 and 12 months were obtained from commercial dealers and were specifically trained for the present experiment. The animals were housed in individual cages and remained in good health during the 7-month training and testing period.

Two stimuli were used: A probe tone at either 0·63, 1·0, 1·6, 2·5, 3·5, or 5·0 kHz, and a series of masker tones located at six or seven different frequencies about each probe tone. The masker and probe tone were generated by separate oscillators, controlled by decade attenuators, combined in an electronic mixer, amplified, and connected to a 29·5 cm speaker (J. B. Lansing, model D120F) mounted in a double-walled sound attenuated chamber. The probe frequency could be gated on or off with an electronic switch (rise/decay time was 50 ms), whereas the masker frequency was presented continuously. A frequency counter and voltmeter monitored the frequency and level of these stimuli. The loudspeaker was located 188 cm above the floor and 48 cm above the parakeet’s head. Free-field sound calibrations were made by placing a 12·5 mm condenser microphone (Brüel and Kjaer, model 4133) in the position of the parakeet’s head and at a o degree angle of incidence to the loudspeaker. The electrical response of the microphone was converted to sound pressure level (SPL) and expressed as dB relative to 20 μPa. Details of the calibration procedure may be found elsewhere (Dooling & Saunders, 1975).

The training steps have been reported in detail elsewhere (Dooling & Saunders, 1975; Saunders et al. 1979a) and the salient features will be presented here. The parakeet was inserted in a tube (slightly larger than its body) so that the head protruded from one end. Thin wire leads were wrapped about the thigh of each leg and electrode paste was applied to insure a good contact with the skin. These electrodes were used to apply shock reinforcement (alternating current, 0·08 mA). The tube was secured in a supporting assembly within the test chamber. Directly in front of the parakeet’s beak was a wire rod and the bird was trained to bite this rod during tonal stimulation in order to avoid reinforcement. The biting response was sensed by a high impedance electronic switch.

Three stages were used to train the animals, and in the first the avoidance response was conditioned. A trial commenced with a tone (usually 2·0 kHz at 80 dB SPL). After 4·0 s, intermittent shock (0·5 s on, 0·5 s off) was paired with the tone until the animal bit the bar. The shock was accompanied by a buzzer which acted as a conditioned reinforcer. The biting response usually occurred within a matter of moments after shock onset. The animal could avoid shock and terminate the tone by responding in less than 4·0 s. Daily test sessions contained 20 trials and always used a 120 s intertrial interval. Acquisition of the avoidance behaviour was rapid and 90% or better correct responses were found after 5–6 days of training (Saunders & Else, 1976).

In the second stage the response was generalized to various frequencies and intensities, the intertrial interval was shortened, and ranged between 5 and 40 s. When the bird agairt performed with 90% or better avoidance behaviours, the threshold measurements began. In this third stage, the stimulus was presented at a single frequency beginning at about 80 dB SPL. If a correct response occurred the tone was then attenuated by 10 dB and another trial was run. Trials continued with the tone being attenuated in 10 dB steps with each correct response. At some SPL the bird failed to detect the stimulus, and the shock, buzzer, and tone were paired simultaneously until a response occurred. The SPL was then increased by 5 dB and a final trial run. The threshold value depended on the response during the final trial and was taken as 2·5 dB less than the SPL of the last correct trial in the series. This modified method of limits (Saunders, 1976) was repeated when thresholds were measured at other frequencies. In order to monitor the birds’ performance, a sham trial was inserted into the series just before the final trial. The timing circuits were cycled as on a regular trial, but neither tone nor shock were presented. The occurrence of a response on these trials was noted. In well-trained animals, where the threshold level was roughly established, the shock was generally turned off during the last three or four trials of the descending series. Performance was then maintained by the conditioned reinforcer (buzzer) alone. On about 5 % of the threshold estimates, however, the shock was left on, but set at a lower level (0·05 mA). This procedure maintained a high level of performance without increasing the spontaneous response rate.

The audibility curve of the four subjects was ascertained at 21 frequencies between 0·2 and 7·0 kHz. A total of four threshold estimates were made for each bird at each frequency. Threshold testing in the quiet was completed at all frequencies prior to the introduction of masking and this insured that the birds were well experienced with the threshold procedure.

Masker frequencies and SPLs were established for each probe tone. Between two and six masker levels were used and these varied from each other in 10 or 20 dB steps. Six or seven masker frequencies were used and these were usually equally divided above and below the probe frequency (F_p). One masker, however, was always the same as the F_p. The masked thresholds were obtained in a systematic way. For. every masker condition a minimum of three thresholds were obtained in each bird. All of the masker conditions (frequencies and levels) were tested at a given probe tone before moving on to another F_p. However, during the daily test sessions each bird performed at a different F_p. Varying the probe frequency among birds was considered useful in reducing experimenter bias, since this made it quite difficult to develop any a priori feel for how the masking data were progressing. We thought this was important since the majority of testing was undertaking by one of us (R.L.P.). Testing continued until the sensory performance of each bird was examined for all Conditions.

Spontaneous responses were also monitored with the sham trials as previously described. The test sessions lasted about 25–35 min and during that time at least three or four thresholds could be measured. As an additional control, whenever maskers at an 80 or 90 dB level were used, a threshold in the quiet was obtaining at the end of the session. A threshold shift from exposure to intense maskers during the session was never observed.

The animals performed well throughout the seven months of testing. The spontaneous response rate during this time remained fairly constant at 12·3 % and between subjects ranged from 9·1 % to 14·7%. The thresholds measured in the quiet, averaged for all birds, appear in Fig. 1 (solid circles) and were identical to those previously reported for parakeets (Dooling & Saunders, 1975; Saunders et al. 1979a). Also included in this figure are four masking curves (open circles) that were obtained at very similar masker SPLs. The dashed line shows the average masker level (54 dB SPL) and the actual levels used were within + 5 dB of this value. These masking curves are plotted in dB relative to 20 μPa in order to show their relation to the audibility curve.

The following points about the masking contours in Fig. 1 can be made using the curve whose peak is at 1·6 kHz as an example. Each entry on this contour represents the threshold of the 1 ·6 kHz probe tone measured in the presence of the indicated masker frequency. The threshold sensitivity of the probe was most drastically shifted when the masker and probe frequency were the same. However, as the masker frequency moved either above or below 1· 6kHz, the amount of masking declined. Indeed, when the masker was set at 0·63, 3·5 or 5·0 kHz the threshold values of the 1·6 kHz probe tone were close to those measured in the quiet. These characteristics of the 1·6 kHz masking contour can be seen in the other curves. In summary, the greatest masking (as indicated by the largest threshold shift) always occurred when the probe and masker tone were the same frequency. As the masker frequency becomes more remote from the probe tone, the threshold sensitivity of the probe begins to approximate that measured in the quiet. The standard deviation at each of the measurement points in these masking contours ranged between 1·9 and 3·8 dB.

The results presented in Fig. 2 summarize the masking contours obtained over all combinations of probe frequency (F_p) and masker level. The parameter associated with each family of curves is the masker SPL, and in each panel this level, in dB, is indicated to the right of the respective contour. The abscissa shows the masker frequency and the ordinate indicates the amount of masking for the probe tone plotted as dB threshold shift relative to the probe-tone quiet threshold.

The masking contours at each probe frequency are remarkably parallel to each other over the range of masking intensities used. Even with masker levels as high as 86 or 90 dB, there was no appreciable change in the shape of the masking contour. The standard deviations averaged over all the data points in Fig. 2 was only 3·1 dB, and this value was similar to the variability seen in threshold measurements in other parakeet experiments (Saunders et al. 1979 a).

Since the masking curves at a given probe-tone frequency were nearly parallel to each other (Fig. 2), the data were collapsed into a common set of masking curves. The was achieved by normalizing the individual curves in the following way : the maximum threshold shift was called zero dB and all other thresholds were measured relative to this. The relative dB values were then averaged over the different masker levels for each probe tone. The 30 dB masker level at 1·6 and 3·5 kHz were not included in this procedure because the number of maskers tested were not sufficient to reveal the exact shape of the masking contour. The results of this operation appear in Fig. 3 and the overall shape of the masking curves, as probe-tone frequency changes, is made clearer.

In order to quantify the shape of these curves in more detail a Q_10dB ratio was calculated. This ratio is obtained by dividing the centre frequency of the filter by the bandwidth of the filter measured at a point 10 dB below the centre frequency. The advantage of this ratio is that it permits relative comparisons of filter sharpness (i.e. frequency selectivity or tuning) between curves whose centre frequencies (i.e. point of maximum masking in the present case) differ from one another. The higher the value of Q_10dB the sharper or more frequency selective is the filtering process. The results of the Q_{10 dB} calculation for each of the six masking curves are included on the figure. The Q_10dB value increases and reaches a peak at 3·5 kHz. This indicates that the masking curve at 3·5 kHz exhibits the greatest frequency selectivity. The 5·0 kHz masking curve shows a lower Q_10dB value meaning that frequency selectivity deteriorates again.

There was also an Orderly relation between the masker level and the amount of probe tone threshold shift, when the masker and probe frequency were the same. This relation is most easily seen when the masked threshold and the masker level are normalized to one another in relative dB. This is accomplished by describing the amount of masking in threshold shift terms and the masker level as dB above the respective probe-tone quiet threshold (i.e. as a sensation level). Fig. 4 plots these data and a regression analysis on the results shows that masking (threshold shift) is highly correlated (r = 0·98) with masker level (sensation level), and that there is slightly less than a one dB change in threshold shift for each dB change in masker intensity.

Vogten (1974) used a similar procedure to measure isointensity masking contours in human listeners. His subjects were only tested with a 10 kHz probe tone but various masker levels were used. We present a comparison between human and parakeet masking contours both obtained at an 80 dB masker level in Fig. 5. These data thus represent threshold performance measured under nearly identical stimulus conditions. The Q_{lo dB} calculations for the two curves reveal that the parakeet was slightly more selective than the human observer at this probe frequency. Furthermore, the parakeet isointensity curve is nearly symmetrical on the high and low frequency sides, while the human curve has a relatively shallow slope on the low frequency side and a much sharper slope on the high frequency side of the curve. The asymmetry of the human isointensity masking contour becomes even more pronounced as the masker level increases. This observation is interesting because the masking contours in the parakeet do not appear to become more or less asymmetrical with changes in masker level (see Fig. 2).

The present data provide further information on the process of frequency resolution in the parakeet and in birds in general. The findings are in agreement with the picture of frequency analysis revealed by other masking procedures in this species, but are somewhat different from the isointensity contours seen in mammalians.

The values of Q_{10 dB} have been calculated for critical ratios, psychophysical tuning curves, evoked response tuning curves, and for the present data, and these results are summarized in Fig. 6. As can be seen the change in frequency selectivity across test frequency is remarkably consistent over the various procedures. All measures of parakeet frequency selectivity are poor in the low frequencies, show greatest selectivity in the frequency region around 3·5 kHz, and again become less selective in the highest frequencies. All the data in this figure were obtained with simultaneous masking procedures.

These data indicate that the frequency resolving power of the parakeet ear is best over a narrow range of frequencies. Moreover, the same frequencies exhibit the most sensitive thresholds and the smallest frequency difference limens. The importance of these observations is that the dominant energy levels in the spectrum of parakeet vocalizations also occur in this frequency range (Dooling & Saunders, 1975). It would thus seem that the parakeet’s auditory system is organized for analysing its own species calls, and this is not so surprising since the highest rate of information transfer would be expected to occur when the receiver and transmitter are optimally tuned to one another.

The essential value of these masking curves is that they describe a system response for the condition of a constant input stimulus. Equally important is the fact that isointensity masking contours are similar to the ‘response area’ curves used to describe the discharge rate of auditory nerve fibres (Rose et al. 1971; Geisler, Rhode & Kennedy. 1974). This similarity is based on the appearance of human masking (Vogten, 1974) and primate auditory nerve (Rose et al. 1971) isointensity contours, and on the underlying assumptions used to interpret these data. The physiological response area curve relates fibre discharge rate to stimulus frequency, when stimulus SPL is held constant. These neural isointensity curves are symmetrical about the characteristic frequency (the frequency to which the fibre responds best) when the stimulus level is low. However, as the level is raised, the shape of the contour becomes increasingly asymmetrical with a shallower slope on the low frequency side and steeper slope on the high frequency side of the curve. An identical result is seen in human masking isopotential contours as (illustrated in Fig. 5 (Vogten, 1974). Furthermore, the auditory nerve discharge rate reflects in part the vigour with which a particular region of the basilar membrane is stimulated. Similarly, the probe-tone threshold is derived, in part, from the sensitivity of a particular region of the basilar membrane. Thus, the neural discharge rate and behavioural threshold are related to one another, since they are both derived from activity at a specific place within the cochlea. It follows that the fibre response rate during stimulation by a constant SPL tone is analogous to the probe-tone threshold during masking by a constant SPL tone. The point to be made by this part of the discussion is that the specific appearance of the isointensity masking contour is, in large part, due to events in the auditory periphery (for a further discussion of this point see Saunders et al. 1979a). In spite of these similarities between the neural and behavioural data, important differences need to be recognized. The behavioural masking contours are defined by events occurring at the highest level of neural integration and thus represent a summation of the processes seen at the single fibre level. Furthermore, the fibre response area curve uses only a single tone, whereas the behavioural area curve uses two stimuli (both occurring simultaneously). The nonlinear effects of two-tone or lateral suppression (Sachs & Kiang, 1968) have been shown to broaden the so called ‘psychophysical’ tuning curve (Moore, 1978; Vogten, 1978 b) when a simultaneous masking procedure is used. It may well be that two-tone suppression also broadens the present masking curves.

The isointensity masking contours in the parakeet differ in several respects from the description of either neural or behavioural isointensity curves in mammals. As masker level increased there was no change in the low or high frequency slope of the curve in the parakeet. Other measures of frequency selectivity in the parakeet indicate that the shape of the filter curve is independent of stimulus level (Saunders et al. 1978b; Saunders & Else, 1976; Saunders, Else & Bock, 1978 c; Saunders, Bock & Fahrbach, 1978a; Saunders et al. 1979a; Saunders, Rosowski & Pallone, 1979b). As noted previously, the shape of human masking contours is very level-dependent (Vogten, 1974). We suggest that the non-linear processes in the auditory system of mammals, which makes the shape of the masking curve dependent on level, do not seem to be active in the same way in the parakeet auditory system. Non-linear processes have been reported for auditory nerve fibre responses in birds (Sachs, Young & Lewis, 1974), but they have not yet been observed in behavioural measures of frequency selectivity in the parakeet.

Perhaps what is most surprising is that the degree of frequency resolution in the parakeet, as described by the present data and those in Fig. 6, is as good as it is. The primary frequency analyser, the basilar papillar, is only 3·2 mm long in this species (compared to the basilar membrane which is 33 mm long in man). Indeed, at the comparison frequency illustrated in Fig. 5, the auditory filter in the parakeet was slightly more selective than in man. It would appear that the parakeet cochlea can perform signal analysis as efficiently as the mammalian cochlea, but only over that narrow range of frequencies which are adaptively significant to the animal (Saunders et al. 1979 a).

The isointensity masking contours add further to the information concerning frequency selectivity in the parakeet. The contours exhibit shape which was independent of level and this was similar to the masking data in parakeets observed by other procedures. This observation was somewhat surprising since it is quite different from the level dependent masking curves observed in mammalian species. We have suggested that non-linear processes in the auditory system, which are most likely responsible for changes in the shape of masking curves with level (Vogten, 1978 a), may not operate in such an obvious way in parakeets. Non-linear processes in birds have been reported in eighth nerve recordings in pigeons (Sachs et al. 1974) and are found in cochlear microphonie recordings of numerous other avian species (Saunders, Coles & Gates, 1973). In the parakeet, however, the behavioural manifestation of these non-linear processes are not yet clear.

Frequency selectivity in this species is very sharp over that narrow range of frequencies which are important for communication. It would be valuable to know whether or not there are any special cochlear adaptations which account for the sort of frequency selective curves seen in Fig. 6.

This research was supported in part by an award from the National Science Foundation (BNS77-26868) and was reported in part at the 97th meeting of the Acoustical Society of America, Cambridge, Mass., June, 1979. The authors appreciate the critical comments of Dr E. M. Relkin and the assistance of Mr Ira Horowitz and Ms Rosemary Osborne. R. L. P. is presently a predoctoral fellow in the Department of Regional Science, University of Pennsylvania.