In vision, colour constancy allows the evaluation of the colour of objects independent of the spectral composition of a light source. In the auditory system, comparable mechanisms have been described that allows the evaluation of the spectral shape of sounds independent of the spectral composition of ambient background sounds. For echolocating bats, the evaluation of spectral shape is vitally important both for the analysis of external sounds and the analysis of the echoes of self-generated sonar emissions. Here, we investigated how the echolocating bat Phyllostomus discolor evaluates the spectral shape of transient sounds both in passive hearing and in echolocation as a specialized mode of active hearing. Bats were trained to classify transients of different spectral shape as low- or highpass. We then assessed how the spectral shape of an ambient background noise influenced the spontaneous classification of the transients. In the passive-hearing condition, the bats spontaneously changed their classification boundary depending on the spectral shape of the background. In the echo-acoustic condition, the classification boundary did not change although the background-and spectral-shape manipulations were identical in the two conditions. These data show that auditory processing differs between passive and active hearing:echolocation represents an independent mode of active hearing with its own rules of auditory spectral analysis.

The colour constancy of the visual system is a well-known perceptual phenomenon that enables the constant perception of an object's colour despite changes in the spectral composition of the light illuminating the object(Smithson, 2005). Humans perceive the colour of a white object still as white and the colour of a red object still as red, independent of illumination by yellowish candle light or greenish neon light.

The timbral constancy of the auditory system(Risset and Wessel, 1982) is an analogous perceptual phenomenon that allows the perception of the spectral shape of a sound (i.e. its timbre or acoustic colour) despite spectral changes caused by the environment (Watkins,1991). The spectral shape of a sound signal is one major component for the identification, grouping and segregation of different auditory signals(Bregman, 1990; Moss and Surlykke, 2001; Griffiths and Warren, 2004). However, the spectral shape of acoustic waves is distorted during propagation in an unpredictable manner (Wiley and Richards, 1978; Lawrence and Simmons, 1982), thus blurring the information content of the sound. The spectrum of a sound reaching the ear is hence the product of the spectrum of the sound source and the spectrum of the surrounding.

Two options exist to cope with such acoustic distortions: either the sender adapts its vocalization behaviour, or the receiver compensates for these distortions. Species of almost all vertebrate classes show the first option,i.e. changes in their vocalization behaviour, e.g. frogs(Feng et al., 2006), birds(Lengagne et al., 1999; Slabbekoorn and Peet, 2003; Brumm, 2004; Slabbekoorn and den Boer-Visser,2006), monkeys (Egnor et al.,2007) and whales (Au et al.,1985; Miller et al.,2000; Foote et al.,2004). Bats, in addition, are generally very adept in adjusting their echolocation calls to changed acoustic conditions(Kalko and Schnitzler, 1993; Gillam et al., 2007). The second option, the timbral constancy of perceived sounds, has only been investigated in humans. Despite the ubiquitous distortions of the spectral envelope of acoustic signals, humans can easily identify them. The auditory system compensates for the spectral characteristics of the environment,resulting in an undistorted perception of the original spectrum of the signal(Risset and Wessel, 1982; Watkins, 1991). When the frequency response of the environment is experimentally manipulated, timbral constancy leads to several perceptual phenomena, such as the `phoneme boundary shift' between intergradient vowels(Watkins, 1991), the `flat spectrum vowel effect' (Summerfield et al., 1987) and the `Zwicker tone'(Zwicker, 1964).

Echolocating bats allow comparison between potentially different mechanisms of auditory spectral processing, since they use both passive and active hearing. The spectral information in both contexts is of great importance and needs to be perceived correctly. Passive hearing is used for vocal communication and rustling prey detection(Esser and Schubert, 1998; Boughman and Wilkinson, 1998; Arlettaz et al., 2001) and we would expect bats to compensate for spectral distortions in the same way as humans. Active hearing, or echolocation, is used for spatial orientation and airborne prey detection (Schnitzler et al., 2003) by listening for the echoes of self-generated ultrasonic calls and evaluating their temporal and spectral properties. For example, bats identify and discriminate objects exclusively on the basis of their different spectral reflection patterns(von Helversen and von Helversen,2003). Bats even separate and generalize the size-invariant spectral reflection pattern of the same object with different sizes from the size-variant spectral reflection pattern(von Helversen, 2004; Simon et al., 2006).

The neuronal processing of auditory information is different in passive hearing and active echolocation and is closely coupled to and directly influenced by the vocalizations during echolocation(Suga and Schlegel, 1972; Schuller, 1979). The behavioural responses to auditory stimuli may thus differ in the same experimental task between passive and active hearing. However, whether this is the case was only investigated in one study on the temporal processing of communication sounds and echoes(Schuchmann, 2006).

In this study, we examined, in two experiments, how bats classify the spectral shape of transient foreground stimuli in relation to a simultaneous noise background with a constant spectral shape. The two experiments differed only in the type of signal that needed to be classified: in the passive-acoustic experiment, the bats had to classify passively presented filtered impulses, which were temporally uncorrelated to their echolocation calls. In the active-acoustic experiment, filtered replica of their recorded echolocation calls were played back in real-time through a loudspeaker, thus generating a virtual echoacoustic object. During both experiments, a continuous noise background was presented, the spectral shape of which was constant during a given trial and could be altered between trials.

During passive hearing, we expected the bats to use the background as a reference for the spectral analysis of the transient foreground stimuli, as humans do (Watkins, 1991). Thus, if the spectral shape of the background was changed, we hypothesized that there would be a shift in the classification as a result of the compensation for the changed environmental transfer characteristics. During echolocation, we hypothesized that there would be no influence of a passively presented constant noise on the spectral classification of echoes, if bats processed echoes differently from passive acoustic signals.

In two separate experiments on passive and active hearing, bats of the species Phyllostomus discolor were trained to classify signals with different spectral shape as containing more energy either in the high or low frequencies. After a training period with two signals that were easily distinguishable by their spectral shape, the bats' spontaneous classification of intermediate test signals was assessed depending on the spectral shape of a noise background.

Animal housing and training procedure

Nine adult, male pale spear-nosed bats, Phyllostomus discolorWagner 1843, were trained in the experiments. They were housed in groups of two to five animals per cage (80 cm×60 cm×80 cm) on a 12 h:12 h dark:light cycle, but were allowed to fly in a large room for several hours each day. Animals had free access to water. On training days, they received food only as a reward in the experiment. Additional food (fruit and mealworms)was given at the weekends. Bats were usually trained daily for 20–30 min on 5 days per week, followed by a 2 days break.

Bats were trained in a two-alternative, forced-choice paradigm to discriminate a highpass filtered training stimulus (+6 dB/octave) from a lowpass filtered training stimulus (–6 dB/octave) during continuous playback of white noise. Bats indicated their decision by crawling into the left or right arm of a Y-shaped maze. Correct decisions were rewarded with banana-pulp; wrong decisions were neither rewarded nor punished. The next trial started when the animal had crawled back to the starting position at the top of the Y-shaped maze. The presented stimulus, and thus the rewarded side,was selected pseudo-randomly (Gellermann,1933).

Data collection of the test conditions began after the bats classified the two training stimuli correctly with >75%. Presentation of training stimuli was continued, serving as controls for a constant classification performance,while the test stimuli were interspersed between these control trials with a probability of 20–40% (Experiment 1) or 20% (Experiment 2). During a test trial, one out of 10 different test stimuli was selected randomly (see the next sections for description of the stimuli). Test trials were always rewarded, independent of the bat's decision. Only the test trials of days with a control performance >75% correct were included into the analysis.

We collected data of three bats in Experiment 1 (passive hearing) and of two bats in Experiment 2 (echolocation). Five animals (two in Experiment 1,three in Experiment 2) did not learn to classify the control stimuli and could thus not be used for testing. Most of these individuals were too inactive and did not explore the setup. This was, for example, the case for Bat 2, which was successfully trained in Experiment 1, but did not cooperate anymore in Experiment 2. Other bats were too active and crawled quickly through the setup without paying enough attention. For example, one bat in Experiment 2 learned the task, but only showed a stable performance once or twice within several weeks.

Experiment 1: passive hearing

Stimuli

A noise background of 45–55 dB SPL re. 20 μPa and 20–100 kHz bandwidth was continuously presented to the bats. The spectral envelope of the noise was either white [i.e. having a flat amplitude spectrum (0 dB/octave),used during training and testing] or pink (i.e. an amplitude spectrum decreasing by –3 dB/octave, test trials only; Fig. 1).

Impulses of 7 μs duration with a flat frequency spectrum were used as foreground stimuli. They were presented at a repetition period of 500 ms,uncorrelated to the bats' echolocation calls. Bats were trained to discriminate impulses that were either high- or lowpass filtered, such that the amplitude of their spectrum either increased or decreased with 6 dB/octave(Fig. 1). After the bats had learned this task, we tested their spontaneous classification of impulses that were filtered by filters having transfer functions of intermediate slopes(±3.6 and ±1.2 dB/octave; Fig. 1). The level of the filtered impulses was roved between ±6 dB to prevent the bats from using loudness cues for classification.

Altogether, we had two training (control) conditions, which were the±6 dB/octave stimuli filters superimposed on white noise, and 10 test conditions, which were the four remaining stimuli filters of ±3.6 and±1.2 dB/octave superimposed on white noise and all six stimuli filters superimposed on pink noise.

Stimulus generation

The impulses (impulse generator HP 8012B, Hewlett-Packard, Palo Alto, CA,USA) and white-noise background (noise generator 1405, Brüel &Kjær, Nærum, Denmark) were separately digitized and continuously filtered with 21-point digital filters (DAP-boards 5200a, Microstar Laboratories, Bellevue, WA, USA; 250 kHz sampling rate). All filters were normalized to their root mean square (r.m.s.) to maintain a constant signal level. The analogue outputs of the DAP-boards were lowpass filtered at 100 kHz(FT 6, Tucker-Davis Technologies, Alachua, FL, USA), attenuated (Crystal 3310,Cirrus Logic, Austin, TX, USA) and then summed together (SM 3, Tucker-Davis Technologies). The combined stimulus was amplified (Yamaha M 35, Yamaha Corp.,Hamamatsu Shizuoka, Japan for Experiment 1 and Rotel RB 960 BX, Rotel, Halle,Germany for Experiment 2) and played back via one ultrasonic loudspeaker (Technics EAS 10 TH 800D; Matsushita Electric Industrial, Osaka,Japan), which was placed in the middle between the two arms of the Y-shaped maze.

Subjects

Data from three bats were collected in Experiment 1, with 23–30 trials per subject and test condition. The number of control trials for white-and pink-noise background, respectively, were 346/354 (Bat 1), 270/268 (Bat 2)and 273/250 (Bat 3).

Experiment 2: echolocation

Stimuli

Instead of impulses, filtered replicas of the bats' echolocation calls were used as foreground stimuli, thus presenting virtual objects reflecting an echo every time the bat was emitting a call. The filters used for the foreground and background were the same as in Experiment 1(Fig. 1). We thus presented again two training/control conditions and 10 test conditions.

Stimulus generation

Echolocation calls were picked up with a microphone (MTG MV301, protection grid off, Microtech Gefell, Gefell, Germany; pre-amplifier 2671, Brüel& Kjær), which was placed on top of the loudspeaker in the middle between the two arms of the Y-shaped maze. Calls were amplified (measuring amplifier 2610, Brüel & Kjær) and bandpass filtered (model 3550 4th order bandpass filter 30–100 kHz, Krohn-Hite, Brockton, MA, USA),before they were processed by the DAP-board as described for Experiment 1 and then played back via the loudspeaker. In addition to normalizing the stimulus filters to their own r.m.s., they were normalized to the r.m.s. of a model call of P. discolor filtered with the respective filter. Virtual object target strength was about –10 dB, which is around the upper bound of large three-dimensional objects relevant for orientation, such as tree trunks (Stilz, 2004). The electronic delay was 3 ms, thus positioning the virtual object about 50 cm behind the loudspeaker. The noise background was generated and filtered as in Experiment 1.

Subjects

Data of two bats were collected with 42 (Bat 4) or 64 (Bat 5) trials per subject and test condition. The number of control trials for white- and pink-noise background, respectively, were 878/916 (Bat 4) and 1416/1439 (Bat 5).

Echolocation call recording and analysis

We recorded echolocation call sequences of the preceding 4 s before decision during almost all trials in Experiment 2 (phase 24, Terratec,Herrenpfad, Germany; 192 kHz sampling rate, 24 bit), altogether 2134 call sequences for Bat 4 and 3457 for Bat 5. We analysed the calls automatically by a custom-written routine (Matlab 7.1, The Mathworks, Natick, MA, USA). Calls were detected as regional maxima above a constant threshold in the lowpass filtered (700 Hz) Hilbert envelope and then extracted form the time signal,containing 5–95% of the total noise-corrected call energy. All calls with a signal-to-noise ratio <30 dB and a peak amplitude <–20 dB FS (decibel full scale) were removed and the remaining calls were checked visually to exclude obvious artefacts, e.g. clicks or external noise. From the five calls with the largest signal-to-noise ratio per sequence (=per trial),we calculated the mean per sequence of six call parameters (best frequency,–20 dB bandwidth and its corresponding lower and upper cut-off frequencies, frequency centroid and fundamental frequency). For further analysis, the sequence means were grouped, either per stimulus or per noise background, to calculate the respective second-order means per stimulus or per noise background.

Analysis and statistics

Perceptual classification boundary

A perceptual classification boundary k was calculated for each animal according to the phoneme boundary used by Tyler et al.(Tyler et al., 1982) and Watkins (Watkins, 1991). The perceptual classification boundary is defined as the filter slope that results in 50% of highpass and lowpass classification by:
\[\ k=m{\times}P-c,\]
(1)
where P is the mean percentage of highpass classifications of all six signals (0–100%). k is defined as 0 for P=50%, which is half of all possible highpass classifications, and k is defined as+6 when the steepest highpass filter (+6 dB/octave) receives only 50% of highpass classification, and the others none (P=8.33%). This yields values of m=–0.144 and c=–7.2.

Roving level simulation

We analysed the possible influence of the roving level on echo classification in Experiment 2 by applying a roving level of –6 to +6 dB to the five selected calls per sequence and calculating their frequency centroid on a logarithmic frequency axis. The frequency centroid divides the amplitude spectrum into two parts with equal energy and can therefore be used as a measure for the frequency distribution.

Statistics

As we were limited to a small number of animals, we used each animal as its own control for the behavioural data. Behavioural classifications of the same foreground stimuli with different noise backgrounds were compared using Fisher's exact test. Behavioural mean data were tested using Student's paired one-sided t-tests, testing the null hypothesis against the one-sided alternative hypothesis that the percentage of highpass classification increases with pink-noise background. Student's two-sided t-tests were used to compare means of the call parameters across backgrounds (Matlab 7.1).

Experiment 1: perceptual compensation during passive hearing

Three bats were successfully trained to discriminate highpass filtered impulses (+6 dB/octave) from lowpass filtered impulses (–6 dB/octave),which were superimposed on a white-noise background (control,χ 2-test, P<0.0001). The open symbols in Fig. 2 show that in these trained conditions, all three bats performed highly reliably, in that highpass filtered impulses were classified as highpass in about 90% of the trials and lowpass filtered impulses were classified as highpass in only about 10% of the trials. With a white-noise background, the bats' spontaneous classification of the test stimuli with intermediate filter slopes depended in general monotonically on the filter slope (solid black symbols in Fig. 2). The perceptual classification boundary (vertical black lines in the lower part of the panels in Fig. 2) is on average at–0.1 dB/octave, and thus very similar to the physical filter boundary of 0 dB/octave. With a pink-noise background, however, the bats classified the same stimuli more often as highpass (red symbols in Fig. 2). Consequently, the perceptual classification boundary shifted towards negative slopes (vertical red lines in the lower part of the panels in Fig. 2). For the individual bats, this downward shift of the perceptual classification boundary amounted to 1.8, 1.0 and 0.8 dB/octave, with a significant mean downward shift of 1.2 dB/octave [Student's one-sided paired t-test, t(2)=5.52, P=0.0156]. Accordingly, the mean difference in highpass classification between pink and white-noise background (black bar in Fig. 2D) is significantly larger than zero [Student's one-sided t-test, t(5)=2.21, P=0.0391]. In summary, all three bats compensated for the spectral shape of the noise background.

Note that the possible shift of the perceptual classification boundary is limited by ceiling effects. These are apparent in the averaged psychometric functions (Fig. 2D) where the degree of highpass classification of signals with a slope of +3.6 or +6 dB/octave did not increase further when the background was changed from white to pink.

Experiment 2: spectral processing during echolocation

In this experiment, the bats were trained to classify the spectral transfer function of a filter, and not the presented sounds themselves. As the filter was excited by the bats' echolocation calls, the perceptual task was to evaluate spectral changes of the perceived echoes, relative to the emitted calls.

Two bats were successfully trained to discriminate highpass filtered echoes(+6 dB/octave) from lowpass filtered echoes (–6 dB/octave), superimposed on white-noise background (control, χ2-test, P<0.0001; open symbols in Fig. 3). As we showed for the passive-acoustic classification of impulses, the spontaneous classification of filters with intermediate spectral slopes and a white-noise background depended monotonically on the filter slope(solid black symbols in Fig. 3). However, in contrast to the passive-acoustic experiment, the classification of the same test filters was not affected by the change from a white to a pink-noise background (red symbols in Fig. 3). The shift of the perceptual classification boundary was positive in Bat 4 and negative in Bat 5, with a non-significant mean shift of –0.3 dB/octave. Accordingly, the mean difference in highpass classification between pink- and white-noise background equals almost zero (black bar, Fig. 3C).

Effect of the roving level

In both experiments, the foreground level was roved to prevent the bats from using loudness cues for classification. In addition to the filter slope,which was the reinforced cue, this roving level also influenced the classification during Experiment 2 (echolocation), but not during Experiment 1(passive hearing). Fig. 4A,Billustrates this effect for the control data of Experiment 2, which are plotted as a function of echo level. This shows that highpass controls were mostly classified as highpass (>90%), with a small decrease in performance for louder echoes. By contrast, lowpass controls were classified correctly at high levels, but incorrectly at low levels with errors of up to 60%. A similar, more pronounced pattern is visible for the test echoes(Fig. 4C,D).

Echolocation calls

P. discolor emits multiharmonic, downward frequency-modulated calls. Calls were very short, having 90% of their energy within less than 0.7 ms (median ∼0.4 ms; Fig. 5A). The mean fundamental frequencies were 18.4 and 19.5 kHz for Bats 4 and 5, respectively (Fig. 5H). The –20 dB bandwidth was around 55 kHz with cut-off frequencies at 36–38 kHz and at 91–93 kHz(Fig. 5C–E), which was about the same for both bats. Inter-individual differences were only found in the energy distribution across frequencies: the frequency centroid of Bat 4 is about 4–5 kHz lower than in Bat 5(Fig. 5F), and its best frequency is about 15 kHz lower than in Bat 5(Fig. 5G). Thus, calls of Bat 4 had most energy in the third harmonic (58–60 kHz), whereas calls of Bat 5 had most energy in the fourth harmonic (75–76 kHz). Hence, both bats performed the spectral classification task with partially different spectral call structures.

The noise background and the different filter slopes only partially influenced the call structure. Bat 4 showed no systematic background-correlated differences in its call parameters. Bat 5 slightly increased the level of the higher harmonics of its calls during the presentation of pink-noise background: the –20 dB bandwidth and the frequency centroid was about 1 kHz higher[Fig. 5D,F, t(3099)=–4.04, P<0.0001], which was caused by an increase in the upper cut-off frequency[Fig. 5E, t(3099)=–4.04, P<0.0001]. Alternatively, it is possible that the bat accomplished this frequency increase by reducing its fundamental frequency by 0.1 kHz [Fig. 5H, t(3099)=3.49, P=0.0005], thus also reducing its fifth harmonic by 0.5 kHz. In consequence, the fifth harmonic moves further down into the pass band of the bat's vocal tract, and thus shifts the frequency centroid upwards.

In both experiments, bats were successfully trained to classify the spectral shape of transient foreground signals that differed in filter slope by 12 dB/octave (Fig. 1). Despite the differing spectral shape of impulses and echolocation calls, both source stimuli were sufficiently broadband for a stable classification of these spectral changes. As the stimulus level was roved, the bats were prevented from using potential loudness cues introduced by the filtering. The bats thus learned to evaluate the filter-generated spectral changes.

Experiment 1: perceptual compensation during passive hearing

With pink-noise background, the bats classified the same passively heard impulses more often as highpass than during the presentation of a white-noise background. This shows that bats employ a compensatory mechanism during passive hearing similar to the one described in humans (e.g. Summerfield et al., 1987; Watkins, 1991). The perceptual compensation in humans is presumably based on the different rate of spectrotemporal variation between two different filters(Furui, 1986; Repp, 1987; van Dijkhuizen et al., 1987; Haggard et al., 1987): while the spectrum of transmission channels is fairly constant over time, the spectrum of natural sound sources varies rapidly. By taking the constant fraction of the overall perceived spectrum as a reference, the auditory system compensates for it and evaluates only relative changes to it. This results in a perceptual whitening of the environmental spectrum and in the perception of the undistorted spectral shape of the transmitted sound.

We found the same compensatory mechanism during passive hearing in bats. When the noise background was filtered with a constant lowpass filter, the bats compensated for this filter and perceived the same transient foreground stimuli as containing more high frequencies. Such a compensation allows animals to perceive the spectral shape of acoustic signals as it had been at the sound source, i.e. to perceive the spectral characteristics of the sound source despite changes in the absolute spectral shape of the acoustic signal that is impinging on the ear.

This timbral constancy of the auditory system is a high-level perceptual process (Watkins, 1991), as is the colour constancy of the visual system(Smithson, 2005); the colour constancy being mediated by cells in area V1 and V4 of the visual cortex(Zeki, 1983; Wachtler et al., 2003). As such, they are likely to be under cognitive control and may be employed only if necessary.

Experiment 2: spectral processing during echolocation

In contrast to passive hearing, the classification of echoes during active hearing was independent of the passively presented noise background. This shows (i) that the spectral processing of echoes differs from the spectral processing of transient passive acoustic stimuli, even in the identical experimental situation, and (ii) that the spectral processing of echoes is independent of simultaneous, passively presented acoustic signals.

In contrast to the auditory processing of passive sounds, the neuronal processing of echoes is coupled to, and influenced by, the vocalisation pattern (Suga and Schlegel,1972; Schuller,1979), for example by opening temporal processing windows(Roverud and Grinnell, 1985). Based on neurophysiological data, it was suggested for the gleaning bat Antrozous pallidus that it processes passively heard prey sounds and actively acquired echoes in two different, parallel auditory pathways(Fuzessery, 1994; Razak and Fuzessery, 2002),which converge later in the auditory cortex(Razak et al., 1999). Behaviourally, it was suggested that the auditory mechanisms for the temporal analysis of spatial echo information are under cognitive control and that they may be employed during passive hearing, but not during echolocation(Schuchmann et al., 2006; Schuchmann, 2006). The current study provides behavioural evidence that spectral processing also differs between passive and active hearing. The reference for the spectral evaluation of echoes is thus not the constant spectral characteristics of external passive acoustic signals, but an internal reference, which can either be an efference copy of the motor signals that generated the last call, or an auditory reference generated from hearing the outgoing call.

In the current experiments, the background was never loud enough to mask the transient foreground. The evaluation of the echo spectral shape independent of ambient background sounds, as it was demonstrated here, does not mean that the bats would not suffer from masking at considerably higher background levels.

Effect of the roving level

In Experiment 2, echo classification was influenced by echo level; softer echoes were more often classified as highpass than louder echoes. Three explanations are conceivable for this: first, the bats might have used residual loudness cues to discriminate the filter slopes of –6 dB/octave and +6 dB/octave, thus rendering the applied roving level ineffective. To test this hypothesis, we filtered the recorded calls with the experimentally used filters and calculated their r.m.s. level. The level of lowpass filtered calls was 3–4 dB lower than the level of highpass filtered calls(Fig. 6B). This finding is in contrast to the experimentally observed level effect(Fig. 5) which showed that softer echoes were more likely to be classified as highpass. Therefore, we reject this hypothesis.

Second, the spectral shape of the echo above hearing threshold might vary with echo level. To test this hypothesis, we applied a roving level of–6 to +6 dB to every recorded call and calculated the frequency centroid of the roved echo above hearing threshold on a logarithmic frequency axis as a measure for the perceived spectral shape. The hearing threshold of P. discolor was taken from Hoffmann et al.(Hoffmann et al., 2008), and for frequencies above 80 kHz from Esser and Daucher[(Esser and Daucher, 1996);magnitudes reduced by 28.5 dB to match the audiogram of Hoffmann et al.(Hoffmann et al., 2008)]. The frequency centroid increased by about 2 kHz for softer echoes(Fig. 6C). In comparison, the filters used for echo generation changed the frequency centroid by about 5 kHz(Fig. 6A), which is about twice as much as caused by the roving level. Frequency cues caused by the roving level should therefore be overruled by the direct spectral changes and are thus not sufficient to explain the influence of the roving level on their own.

Third, timbre, pitch and loudness are auditory perceptual dimensions that mutually interact in a complex non-linear fashion. For example the reaction time for stimulus classification in one of these dimensions depends in non-reciprocal ways on the other dimensions(Melara and Marks, 1990), with high pitch and high timbre both corresponding to loud signals, but high timbre corresponding to low pitch. In echolocation, small targets give more high frequent echo content than larger ones. Therefore, the bats may tend to classify faint targets similarly to high-pass filtered ones.

In summary, the joined effects of the bats perceptual associations between different stimulus qualities and the spectral composition of the echolocation calls combined with the bats' absolute hearing threshold can at least qualitatively explain the level dependence of the behavioural performance in Experiment 2. Note that the shape of the psychometric functions did not change if only a part of the trials was analysed, e.g. all trials with a roving level above 0 dB. Thus, the observed effect of the roving level does not confound the main result, i.e. the lack of influence of background spectral shape on echo spectral classification.

Echolocation calls

In addition to perceptual compensatory mechanisms, which are used during passive hearing, bats could compensate for spectral distortions during echolocation by adapting the call spectrum to the spectral shape of passive acoustic signals. Bats constantly adjust their signals to meet changing perceptual requirements. For example, call bandwidth is changed during prey interception (Schnitzler et al.,2003), presumably to increase object resolution(Siemers and Schnitzler, 2004; Boonman and Ostwald, 2007), and depending on the surroundings (Kalko and Schnitzler, 1993) and the calls of conspecifics(Gillam et al., 2007). If bats compensated for the lowpass characteristic of the passively presented noise background, they would need to shift the frequency centroid of their calls to lower frequencies. However, every consistent effect we found was in the opposite direction, i.e. with a pink-noise background, the frequency centroid was shifted to higher frequencies. These findings do not support the compensation hypothesis. Instead, this is further evidence that echo spectral processing is not influenced by passive acoustic stimuli. Passive and active hearing thus represent two different and separate modes of operation of the auditory system, where the spectral shape of passive sounds does not influence the evaluation of signals during active hearing.

The presented data show that the bat Phyllostomus discolor adapted the processing of passive acoustic signals to compensate for spectral distortions. Communication signals, or other passively heard sounds, can thereby be corrected for spectral changes generated by the environmental transfer function. Such changes are discernible from the constant spectrotemporal part of the same or other ongoing signals, such as ambient sounds, transmitted through the same transmission channel. By contrast, when the bats had to classify self-generated echoes instead of passively presented transient signals in exactly the same experimental paradigm, their classification was independent of the passively heard noise background. The spectral analysis of transient stimuli during echolocation, i.e. spectral echo analysis, thus recruits different auditory processing mechanisms than the spectral evaluation of passively heard sounds. The data provide psychophysical evidence for the hypothesis that bats switch to an independent mode of auditory processing during echolocation, which does not necessarily obey to the general rules of auditory spectral analysis. Thus, during echolocation,spectral echo analysis is not confounded by the spectral shape of faint ambient sounds.

We thank Nadine Becker, Uwe Firzlaff, Sven Schörnich, Björn M. Siemers and two anonymous reviewers for very helpful comments that considerably improved an earlier version of this manuscript. This work was supported by a scholarship of the DFG graduate program 1091 `Orientation and Motion in Space' to H.R.G. The experiments in this study were non-invasive and therefore did not require experimentation approval. Permission to keep and breed Phyllostomus discolor in fulfillment of § 11, Abs. 1 TierSchG has been given to the Dept. Biologie II der Ludwig-Maximilians-Universität München by the Landratsamt München, dated April 20, 2005.

Arlettaz, R., Jones, G. and Racey, P. A.(
2001
). Effect of acoustic clutter on prey detection by bats.
Nature
414
,
742
-745.
Au, W. W. L., Carder, D. A., Penner, R. H. and Scronce, B. L. (
1985
). Demonstration of adaptation in beluga whale echolocation signals.
J. Acoust. Soc. Am.
77
,
726
-730.
Boonman, A. and Ostwald, J. (
2007
). A modeling approach to explain pulse design in bats.
Biol. Cybern.
97
,
159
-172.
Boughman, J. W. and Wilkinson, G. S. (
1998
). Greater spear-nosed bats discriminate group mates by vocalizations.
Anim. Behav.
55
,
1717
-1732.
Bregman, A. S. (
1990
).
Auditory Scene Analysis: The Perceptual Organization of Sound
. Cambridge:MIT Press.
Brumm, H. (
2004
). The impact of environmental noise on song amplitude in a territorial bird.
J. Anim. Ecol.
73
,
434
-440.
Egnor, S. E., Wickelgren, J. G. and Hauser, M. D.(
2007
). Tracking silence: adjusting vocal production to avoid acoustic interference.
J. Comp. Physiol. A
193
,
477
-483.
Esser, K. H. and Daucher, A. (
1996
). Hearing in the FM-bat Phyllostomus discolor: a behavioral audiogram.
J. Comp. Physiol. A
178
,
779
-785.
Esser, K. H. and Schubert, J. (
1998
). Vocal dialects in the lesser spear-nosed bat Phyllostomus discolor.
Naturwissenschaften
85
,
347
-349.
Feng, A. S., Narins, P. M., Xu, C. H., Lin, W. Y., Yu, Z. L.,Qiu, Q., Xu, Z. M. and Shen, J. X. (
2006
). Ultrasonic communication in frogs.
Nature
440
,
333
-336.
Foote, A. D., Osborne, R. W. and Hoelzel, A. R.(
2004
). Whale-call response to masking boat noise.
Nature
428
,
910
.
Furui, S. (
1986
). On the role of spectral transition for speech perception.
J. Acoust. Soc. Am.
80
,
1016
-1025.
Fuzessery, Z. M. (
1994
). Response selectivity for multiple dimensions of frequency sweeps in the pallid bat inferior colliculus.
J. Neurophysiol.
72
,
1061
-1079.
Gellermann, L. W. (
1933
). Chance orders of alternating stimuli in visual discrimination experiments.
J. Genet. Psychol.
42
,
206
-208.
Gillam, E. H., Ulanovsky, N. and McCracken, G. F.(
2007
). Rapid jamming avoidance in biosonar.
Proc. R. Soc. Lond. B Biol. Sci.
274
,
651
-660.
Griffiths, T. D. and Warren, J. D. (
2004
). What is an auditory object?
Nat. Rev. Neurosci.
5
,
887
-892.
Haggard, M. P., Trinder, J. R., Foster, J. R. and Lindblad,A.-C. (
1987
). Two-state compression of spectral tilt:individual differences and psychoacoustical limitations to the benefit from compression.
J. Rehabil. Res. Dev.
24
,
193
-206.
Hoffmann, S., Baier, L., Borina, F., Schuller, G., Wiegrebe, L. and Firzlaff, U. (
2008
). Psychophysical and neurophysiological hearing thresholds in the bat Phyllostomus discolor.
J. Comp. Physiol. A
194
,
39
-47.
Kalko, E. K. V. and Schnitzler, H.-U. (
1993
). Plasticity in echolocation signals of European pipistrelle bats in search flight: implications for habitat use and prey detection.
Behav. Ecol. Sociobiol.
33
,
415
-428.
Lawrence, B. D. and Simmons, J. A. (
1982
). Measurements of atmospheric attenuation at ultrasonic frequencies and the significance for echolocation by bats.
J. Acoust. Soc. Am.
71
,
585
-590.
Lengagne, T., Aubin, T., Lauga, J. and Jouventin, P.(
1999
). How do king penguins (Aptenodytes patagonicus)apply the mathematical theory of information to communicate in windy conditions?
Proc. R. Soc. Lond. B Biol. Sci.
266
,
1623
-1628.
Melara, R. D. and Marks, L. E. (
1990
). Interaction among auditory dimensions: timbre, pitch, and loudness.
Percept. Psychophys.
48
,
169
-178.
Miller, P. J. O., Biassoni, N., Samuels, A. and Tyack, P. L.(
2000
). Whale songs lengthen in response to sonar.
Nature
405
,
903
.
Moss, C. F. and Surlykke, A. (
2001
). Auditory scene analysis by echolocation in bats.
J. Acoust. Soc. Am.
110
,
2207
-2226.
Razak, K. A. and Fuzessery, Z. M. (
2002
). Functional organization of the pallid bat auditory cortex: emphasis on binaural organization.
J. Neurophysiol.
87
,
72
-86.
Razak, K. A., Fuzessery, Z. M. and Lohuis, T. D.(
1999
). Single cortical neurons serve both echolocation and passive sound localization.
J. Neurophysiol.
81
,
1438
-1442.
Repp, B. H. (
1987
). On the possible role of auditory short-term adaptation in perception of the prevocalic [m]-[n]contrast.
J. Acoust. Soc. Am.
82
,
1525
-1538.
Risset, J.-C. and Wessel, D. L. (
1982
). Exploration of timbre by analysis and synthesis. In
The Psychology of Music
(ed. D. Deutsch), pp.
25
-58. San Diego: Academic Press.
Roverud, R. C. and Grinnell, A. D. (
1985
). Echolocation sound features processed to provide distance information in the CF/FM bat, Noctilio albiventris: evidence for a gated time window utilizing both CF and FM components.
J. Comp. Physiol. A
156
,
457
-469.
Schnitzler, H.-U., Moss, C. F. and Denzinger, A.(
2003
). From spatial orientation to food acquisition in echolocating bats.
Trends Ecol. Evol.
18
,
386
-394.
Schuchmann, M. (
2006
).
Spatial echo suppression and echo-acoustic object normalization in echolocating bats
. Dissertation, Ludwig-Maximilians-Universität München,Germany.
Schuchmann, M., Hübner, M. and Wiegrebe, L.(
2006
). The absence of spatial echo suppression in the echolocating bats Megaderma lyra and Phyllostomus discolor.
J. Exp. Biol.
209
,
152
-157.
Schuller, G. (
1979
). Vocalization influences auditory processing in collicular neurons of the CF-FM-bat, Rhinolophus ferrumequinum.
J. Comp. Physiol. A
132
,
39
-46.
Siemers, B. M. and Schnitzler, H.-U. (
2004
). Echolocation signals reflect niche differentiation in five sympatric congeneric bat species.
Nature
429
,
657
-661.
Simon, R., Holderied, M. W. and von Helversen, O.(
2006
). Size discrimination of hollow hemispheres by echolocation in a nectar feeding bat.
J. Exp. Biol.
209
,
3599
-3609.
Slabbekoorn, H. and Peet, M. (
2003
). Birds sing at a higher pitch in urban noise.
Nature
424
,
267
.
Slabbekoorn, H. and den Boer-Visser, A. (
2006
). Cities change the songs of birds.
Curr. Biol.
16
,
2326
-2331.
Smithson, H. E. (
2005
). Sensory, computational and cognitive components of human colour constancy.
Philos. Trans. R. Soc. Lond. B Biol. Sci.
360
,
1329
-1346.
Stilz, P. (
2004
).
Akustische Untersuchungen zur Echoortung bei Fledermäusen
. PhD thesis,Eberhard Karls Universität Tübingen, Germany.
Suga, N. and Schlegel, P. (
1972
). Neural attenuation of responses to emitted sounds in echolocating bats.
Science
177
,
82
-84.
Summerfield, Q., Sidwell, A. and Nelson, T.(
1987
). Auditory enhancement of changes in spectral amplitude.
J. Acoust. Soc. Am.
81
,
700
-708.
Tyler, R. S., Summerfield, Q., Wood, E. J. and Fernandes, M. A. (
1982
). Psychoacoustic and phonetic temporal processing in normal and hearing-impaired listeners.
J. Acoust. Soc. Am.
72
,
740
-752.
van Dijkhuizen, J. N., Anema, P. C. and Plomp, R.(
1987
). The effect of varying the slope of the amplitude-frequency response on the masked speech-reception threshold of sentences.
J. Acoust. Soc. Am.
81
,
465
-469.
von Helversen, D. (
2004
). Object classification by echolocation in nectar feeding bats: size-independent generalization of shape.
J. Comp. Physiol. A
190
,
515
-521.
von Helversen, D. and von Helversen, O. (
2003
). Object recognition by echolocation: a nectar-feeding bat exploiting the flowers of a rain forest vine.
J. Comp. Physiol. A
189
,
327
-336.
Wachtler, T., Sejnowski, T. J. and Albright, T. D.(
2003
). Representation of color stimuli in awake macaque primary visual cortex.
Neuron
37
,
681
-691.
Watkins, A. J. (
1991
). Central, auditory mechanisms of perceptual compensation for spectral-envelope distortion.
J. Acoust. Soc. Am.
90
,
2942
-2955.
Wiley, R. H. and Richards, D. G. (
1978
). Physical constraints on acoustic communication in atmosphere: implications for evolution of animal vocalizations.
Behav. Ecol. Sociobiol.
3
,
69
-94.
Zeki, S. (
1983
). Colour coding in the cerebral cortex: the reaction of cells in monkey visual cortex to wavelengths and colours.
Neuroscience
9
,
741
-765.
Zwicker, E. (
1964
). Negative afterimage in hearing.
J. Acoust. Soc. Am.
36
,
2413
-2415.