The relationship between basilar-membrane (BM) vibrations and the discharges of auditory-nerve fibers (ANFs) has been most systematically explored in chinchilla, especially at the cochlear site with characteristic frequency (CF) of 9–10 kHz. For this site, BM and ANF responses to similar stimuli are available from healthy cochleae in different groups of animals (Robles et al. 1986; Ruggero and Rich 1983, 1987; Ruggero et al. 1990, 1996b) and, exceptionally, also for the same cochleae of 2 individuals (Narayan et al. 1998). These data sets have permitted investigators to draw a number of conclusions regarding the transformations that acoustic signals undergo between mechanical vibrations and the generation of spike trains in the auditory nerve (Ruggero et al. 2000). However, a crucial gap remains in our knowledge of basal regions of the chinchilla cochlea: the phases of responses at CF of high-CF ANFs are not known at all because of the weakness of phase locking at high frequencies (Johnson 1980; Palmer and Russell 1986). Now we bridge that gap, taking advantage of the ability of 2nd-order Wiener kernel analysis to extract timing information even in the absence of phase locking.

The present study complements a companion paper (Recio-Spinoso et al. 2005), which described 1st- and 2nd-order Wiener kernels computed from the responses to white Gaussian noise of chinchilla ANFs. Here we validate Wiener analysis by comparing responses to tones, clicks, and noise with predictions based on the Wiener kernels. First we show that the responses to tones and the Wiener kernels of responses to noise have similar frequency tuning, regardless of ANF CF. Second, we exploit the residual phase locking in responses to tones of high-CF ANFs to show a satisfactory match between the timing of responses to tones and noise, again regardless of CF. Third, we show that ANF responses to “frozen” noise are well predicted by the zeroth-, 1st-, and 2nd-order Wiener kernels (h₀, h₁, and h₂), jointly. Finally, we use near-CF group delays from Wiener kernels and signal-front delays from responses to intense rarefaction clicks to estimate the synaptic and neural delays interposed between spike generation and mechanical vibrations of the chinchilla cochlea and to construct maps of BM delay as a function of distance throughout the length of the chinchilla cochlea, a first for amniotic species.

The auditory nerve was approached superiorly after craniotomy and aspiration of part of the cerebellum. Capillary-glass microelectrodes (filled with 3 M NaCl or KCl solutions, impedance 20–70 MΩ) were positioned under visual control through an operation microscope and were advanced into the nerve by means of a remotely controlled hydraulic positioner.

Gaussian white noise was produced using an analog device (General Radio 1381), a Tucker Davis WG1, or, exceptionally (when 2nd-order kernels were not computed), by sampling from a read-only memory containing 8,192 16-bit words. In all cases, the bandwidth of the noise stimulus was reduced to 15 kHz by analog filtering. The levels of the noise stimuli were usually selected so that they elicited rate responses slightly higher than spontaneous activity. Noise levels at CF are expressed as spectrum level (in dB SPL/Hz) or total pressure (in dB SPL) in an equivalent rectangular bandwidth computed individually for each ANF [see METHODS and Fig. 16 of companion paper (Recio-Spinoso et al. 2005) for details].

Noise stimuli were delivered either continuously (duration range: 40 s to 18 min; duration average: 3.75 min) or in multiple presentations of the same waveform (frozen noise; duration: 20, 40, or 100 ms, with duty cycle 1:2, presented with level identical to that of the continuous noise). During continuous tone or noise stimulation, both the stimulus and the evoked spike signals were recorded on a stereo digital audiotape recorder (Sony DTC-690: 16-bit resolution, sampling rate 44.1 or 48 kHz) for later analysis. The stimulus waveform was digitized with 16-bit resolution (TDT AD1) and stored in the computer, which also stored the neural responses as a series of spike times (measured at their initial edge with a TDT ET1) with 1-μs resolution. The frozen noise samples were captured by digitization from the analog noise source and were later presented by a 16-bit D/A converter (TDT DA3-2).

(1)

(2)

where N₀ is the average (i.e., DC) response rate (spikes/s) and ϕ_i is the phase angle of the ith spike (0 ≤ ϕ_I ≤ 2π).

To assess the statistical significance of phase locking of responses to tones (VS_tones), the statistic 2nVS² was computed, where n is the number of spikes (Mardia and Jupp 2000). Except where explicitly noted, 2nVS² > 10.6 for all of the tone data reported here. With 2nVS² > 10.6, phase locking is significant with P < 0.01.

FIG. 6 Wiener-kernel predictions of responses to frozen noise for a low-CF ANF. Averaged responses to 256 repetitions of a frozen noise sample (illustrated at top) are indicated by peristimulus time histograms (PSTHs; black traces in A–D). Response predicted (more ...)

FIG. 8 Wiener-kernel predictions of ANF responses to frozen noise. Averaged responses to many presentations of a frozen noise sample (illustrated at top) are indicated by PSTHs (black traces). ANFs had CFs between 0.4 and 11.5 kHz. Responses predicted jointly (more ...)

The ability of the kernels to account for responses to frozen noise was quantified using the coefficient of determination (COD)

(3)

where N is the number of time bins in the measured response (the PSTH) and the predicted response, p_i is the predicted instantaneous rate in time bin i (e.g., red traces in Figs. 6–8), and m_i is the measured instantaneous rate in time bin i (e.g., black traces in Figs. 6–8). The COD is expressed in % throughout the paper.

We tested the validity of Wiener-kernel analysis by comparing responses to tones and frozen noise with response predictions derived from the Wiener kernels computed for the same ANFs. In addition, we compared cochlear delays derived from the Wiener kernels and from ANF responses to intense clicks with direct measurements of the corresponding delays in BM responses to tones, noise, and clicks.

FIG. 1 Frequency tuning of responses to tones and noise recorded in the same low–characteristic frequency (CF) auditory-nerve fibers (ANFs). Responses to tone pips are represented by their frequency–threshold tuning curves (dotted lines). (more ...)

For ANFs with BF >2 kHz, the signal-to-noise ratios of the h₁s decreased steadily as a result of the loss of phase locking, impeding the reliable measurement of frequency tuning in high-CF ANFs. For such ANFs, reliable measurements of frequency tuning in responses to noise were derived from h₂-FSVs (Fig. 2). Figure 2 compares frequency tuning in responses to tones and noise of high-CF ANFs. In general, the magnitude spectra of h₂-FSVs (continuous line) were very similar to the FTCs of responses to tones (dotted line) within about 20–30 dB of CF threshold.

FIG. 2 Frequency tuning of responses to tones and noise recorded in the same high-CF ANFs. Responses to tone pips are represented by the frequency–threshold tuning curves (dotted lines). Responses to noise are represented by the magnitudes of the Fourier (more ...)

Figure 3 explores the strength of phase locking in responses to noise and tones as a function of frequency. The thick dashed line indicates the trend line for a measure of reverse-correlation (revcor) magnitude, VS_noise [see Fig. 12 in the companion paper (Recio-Spinoso et al. 2005)], which is analogous to vector strength for responses to tones (Goldberg and Brown 1969). This trend line is also consistent with the vector strengths for responses to noise (VS_noise, diamonds) and tones (VS_tones, circles and squares) recorded in the same ANFs (phases plotted in Fig. 5B). Note that good matches between noise and tone data extend to at least 10.5 kHz. For comparison with the chinchilla data, Fig. 3 includes vector-strength trend lines for responses to tones in guinea pig (dashed line) and cat (dotted line) ANFs (from data of Palmer and Russell 1986 and Johnson 1980, respectively, summarized in Weiss and Rose 1988). Except for probable differences in corner frequencies, vector strength in all 3 species can be simply described as a low-pass function with high-frequency slope of roughly −18 dB/octave. The corner frequency for chinchilla ANFs is intermediate between those for cat ANFs (highest) and guinea pig (lowest).

FIG. 3 Strength of phase locking as a function of CF in ANF responses to noise and tones. Thick dashed line indicates the trend of the magnitude of the 1st-order cross-correlograms (revcors) normalized to stimulus amplitude [from Fig. 12 of the companion (more ...)

FIG. 5 Phases of responses to tones and noise recorded in the same ANFs. Lines indicate the phases of responses to noise, as a function of frequency. Filled symbols indicate the phase at BF. Open symbols indicate responses to tones of the same ANFs. A: comparisons (more ...)

The correspondence between Wiener kernels and the FTCs for responses to tones is further explored in Fig. 4. Figure 4A shows that the CFs for responses to tones were nearly the same as the best frequencies (BFs) of responses to noise measured from revcors, h₁s and h₂s in the same ANFs. [We distinguish between CF, the frequency that yields the most sensitive responses at threshold levels in normal adult cochleae, and BF, the frequency of peak sensitivity. Each ANF has a unique CF but has various BFs that depend on cochlear maturity (Overstreet et al. 2002) or health and stimulus level.] It is apparent from Fig. 4A that, in spite of decreasing signal-to-noise ratio as a function of increasing BF, BFs could be reliably determined from windowed h₁s [see Fig. 10B of companion paper (Recio-Spinoso et al. 2005)] in many ANFs with BF as high as 10 kHz.

FIG. 4 Frequency tuning measured with tones and with noise. A: peak frequencies [best frequencies (BFs)] of the windowed h1s [closed circles; see Fig. 10B of the companion paper (Recio-Spinoso et al. 2005)] and the h2-FSVs (open circles) are plotted against (more ...)

Figure 4B compares the sharpness of tuning of responses to tones and noise, measured as Q_10dB (CF or BF divided by the bandwidth of tuning at +10 dB re CF or BF threshold). In general, responses to noise and tone stimuli revealed frequency tuning that varied similarly with CF. However, the sharpness of tuning of many high-CF ANFs was underestimated by the h₁s (closed circles in Fig. 4B) as a result of their poor signal-to-noise ratio [Fig. 2; see also Fig. 12 of companion paper (Recio-Spinoso et al. 2005)].

Figure 5A compares the phases of responses to tones (open circles) and noise (lines, from the Fourier transforms of revcors) for several low-CF ANFs. It is evident that noise and tone stimulation yielded very similar response phases, both at CF (filled circles) and at higher and lower frequencies.

Timing information from the near-CF responses of high-CF ANFs (Fig. 5B) was obtained from the h₂-FSVs [see red traces of Figs. 4A and 9, and color traces of Fig. 14A of companion paper (Recio-Spinoso et al. 2005)]. In 7 of these ANFs, responses to tones were also measured at CF and/or at other frequencies. [The strength of phase locking (vector strength) for each of the responses to tones of the ANFs of Fig. 5B is indicated in Fig. 3.] All but one of the responses to tones of the 4 ANFs with CFs <6 kHz exhibited highly significant phase locking (2nVS² >10.6; P < 0.005) but the ANFs with CFs >6 kHz were not significantly phase locked. Nevertheless, Fig. 5B shows that the phases of responses to tones with frequency >6 kHz (open symbols) were very similar to the phases of responses to noise (lines, from the Fourier transforms of the h₂-FSVs) at the same frequencies. On average, the phases of responses to tones of the ANFs of Fig. 5B differed from the phases of the corresponding responses to noise, at the same frequencies and comparable stimulus levels, by only 21 ± 20° (n = 7).

FIG. 9 Quantitative estimates of the quality of predictions of responses to frozen noise based on Wiener-kernel analysis. Quality of the predictions of responses from 38 ANFs from 4 chinchillas is quantified using the COD (expressed as a percentage; see Eq. (more ...)

To summarize: 1st- and 2nd-order Wiener kernels of responses to noise correctly predicted the frequency tuning (Figs. 1–4) and timing characteristics (Fig. 5) of responses to tones, regardless of ANF CF.

The black traces of Fig. 6, A–D show the PSTH for the responses of a representative low-CF ANF to many repetitions of the frozen noise. The histogram consisted of an oscillation with periodicity corresponding to CF (109 Hz). The responses predicted by h₀ and h₁ individually are shown in Fig. 6A (green and blue traces, respectively). It is apparent that the waveform predicted by h₁ accounts for much of the response, its major deficiency being that it is not rectified. The absence of rectification is ameliorated by including h₀ (Fig. 6A, green trace) in the prediction (Fig. 6C, blue trace). Figure 6B shows the individual contribution of h₂. When this contribution is added to that of h₀ and h₁, the predicted response (Fig. 6D, red trace) improves further and closely approximates the actual response histogram.

The panels of Fig. 7 show the PSTH (black trace) for responses of a representative high-CF ANF to many repetitions of the frozen noise. In contrast with the responses of low-CF fibers to the same stimulus (Fig. 6), the fine structure of the histograms of Fig. 7 is unrelated to CF. The fine structure probably bears a relation to the average frequency of the random amplitude modulation expected from a narrow bandpass filter stimulated by white noise (Rice 1954; Ruggero 1973). The responses predicted by h₀ and h₂ are shown individually in Fig. 7A (green and red traces, respectively). Their joint prediction (red trace in Fig. 7B) accounts for most of the features of the histogram.

FIG. 7 Wiener-kernel predictions of responses to frozen noise for a high-CF ANF. Averaged responses to 512 repetitions of a frozen noise sample (illustrated at top) are indicated by PSTHs (black traces in A and B). Responses predicted individually by h0 and (more ...)

The ability of the kernels to account for responses to frozen noise was quantified using the COD (see Eq. 3 in METHODS). The COD equals 100% for a perfect prediction. The CODs for the predictions based on the kernels individually or jointly are indicated in red in Figs. 6–8. [The initial 7 ms of the responses were excluded from the computations of the CODs (see blue bars in Fig. 8) because the Wiener kernels, inasmuch as they are based on well-adapted ANF responses during continuous noise stimulation, cannot predict the response onsets.] In the case of the ANF of Fig. 6, with the lowest CF in our sample, the CODs for the joint predictions of h₀, h₁, and h₂ averaged 83%. In other words, the 3 kernels jointly predicted 83% of the features of the response histograms. In the case of the high-CF ANF of Fig. 7, h₀ and h₂ jointly predicted 85% of the measured response. [Note that the sum of the individual CODs (i.e., those computed separately for h₀, h₁, or h₂) is not the same as the COD for their joint prediction.]

Figure 8 compares the measured responses to frozen noise (black traces) and the responses predicted by the Wiener kernels (red traces) for 10 ANFs with widely ranging CFs (392 Hz–11.5 kHz). The responses to the frozen noise differed between low- and high-CF ANFs in that the carrier was absent from the latter, but all responses, regardless of CF, were well matched by predictions based on the Wiener kernels.

Figure 9 summarizes the quality of predictions based on the kernels for the entire data set of 38 ANFs stimulated with frozen noise. Triangles and open and closed circles in Fig. 9A indicate, respectively, the CODs for the predictions based individually on h₀, h₁, and h₂. In the case of ANFs innervating apical regions of the cochlea, h₀, h₁, and h₂ contributed roughly equally, between 10 and 40%, to the overall prediction of the response histograms. The contributions of h₀ varied widely in high-CF ANFs and were, on average, marginally larger than those in low-CF ANFs. The contributions of the h₁s were relatively constant in low-CF ANFs but insignificant in high-CF ANFs (dashed line). The contributions of h₂ grew monotonically as a function of increasing CF (dotted line). For most high-CF ANFs, the contributions of h₂ were dominant, typically accounting for ≥50% of the responses.

The CODs for the joint predictions of the 3 kernels are shown in Fig. 9B (closed circles). On average, h₀, h₁, and h₂ accounted jointly for 77.4 ± 9.7% (n = 38) of the responses of chinchilla ANFs, regardless of CF (dashed line). For low-CF ANFs (CF <2.7 kHz), the CODs for the 1st- and 2nd-order predictions averaged 65.3 and 75.1%, respectively. For high-CF ANFs (CF >2.7 kHz), the CODs for the 1st- and 2nd-order predictions were 45.8 and 78.6% on average. In other words, the 2nd-order predictions improved the 1st-order predictions by amounts that grew as a function of increasing CF.

Interestingly, it was possible to obtain fairly accurate predictions of the responses to frozen noise by using h₀s and h₁s jointly with severely reduced versions of h_2, constructed on the basis of either the h₂-FSV alone or the singular vectors with 1st and 2nd ranks. Figure 9C shows that the predictions based on the FSV alone (upward triangles) accounted for 71.6% of the response features (i.e., only slightly worse than the predictions based on the complete h₂s) and predictions based on reduced h₂s constructed from the 2 highest ranked singular vectors (i.e., 1st and 2nd ranks; downward triangles) were similar. However, the former (reduced h₂s based on the FSVs alone) yielded better predictions than the latter for low-CF ANFs and worse predictions for high-CF ANFs. These results are consistent with the finding that the singular vectors of rank other than 1st for low-CF ANFs and ranks other than 1st or 2nd for high-CF ANFs represent spurious features accompanying poor h₂ signal-to-noise ratios [Fig. 7B of companion paper (Recio-Spinoso et al. 2005)].

Comparisons between phases of h₁s and responses to tones of low-CF ANFs have not been carried out previously. The present results for chinchilla (Fig. 5A) demonstrate that the phases and group delays of h₁s and responses to tones are similar when measured in the same ANFs.

FIG. 13 Onset latencies and near-CF group delays of responses of ANFs and BM sites in the chinchilla cochlea. A: dashed line is the central tendency for the onsets of chinchilla ANF responses to intense rarefaction clicks (τfront from Fig. 11 (more ...)

With but a single (and probably questionable) exception (Teich et al. 1993), phase-locked responses to tones with frequency as high as 10 kHz (e.g., Fig. 5B) have not been previously demonstrated in ANFs in any mammalian species. Nevertheless, Fig. 3 shows that the rate of decay of phase-locking strength as a function of increasing frequency in chinchilla is roughly consistent with previous results in guinea pig and cat. It has been suggested that residual phase locking for high-frequency stimuli provides significant temporal information and, in humans, plays an important role in encoding frequencies at least ≤10 kHz (Heinz et al. 2001).

Wickesberg et al. (1984) were the first to use h₂s for the study of auditory neurons in the cat cochlear nucleus. They estimated the quality of predictions of responses to noise on the basis of h₀, h₁, and h₂. Even when conditions were optimal, as illustrated in their Fig. 2 (Wickesberg et al. 1984), the 3 kernels jointly predicted <50% of the responses of low-CF neurons and “… the addition of the second kernel improved a prediction based on the zero- and first-order kernels, but not by very much.” In contrast, our general finding for chinchilla low-CF ANFs is that 2nd-order predictions (based on the 3 kernels) almost always improved the 1st-order predictions (based solely on h₀ and h₁).

The single previous use of h₂s to study mammalian ANFs (in gerbil; Lewis et al. 2002) did not evaluate systematically the quality of response predictions based, either individually or jointly, on h₀, h₁, and h₂, apparently because the h₂s measured in gerbil were excessively contaminated by noise (Recio-Spinoso et al. 2005; see also p. 220 of Lewis et al. 2002).

FIG. 10 Onset latencies of ANF responses to rarefaction and condensation clicks. Latencies τ were obtained from compound PSTHs (binwidth: 40 μs) of responses to intense (peak: about 104 dB SPL) clicks [M. A. Ruggero, N. C. Rich and A. N. Temchin, (more ...)

Figure 10B presents the onset latencies of responses to intense condensation clicks. The trend line, obtained by adding 0.5BF periods to the fit function for intense rarefaction clicks (i.e., solid line in Fig. 10A), satisfactorily matches the latencies to condensation clicks for BFs >350 Hz but overestimates those latencies for BFs <350 Hz. This we attribute to the presence of “glides,” from high to low frequencies, at the onset of the Wiener kernels of ANFs with BFs <350 Hz [see Fig. 17A of companion paper (Recio-Spinoso et al. 2005)].

FIG. 11 Filter delay of low-level basilar-membrane (BM) responses. BM filter delays were obtained by subtracting signal-front delays (i.e., the latencies of ANF responses to intense rarefaction clicks) from the near-BF group delays of Wiener kernels of (more ...)

The insignificance of h₂s in BM vibrations indicate that these contain only small amounts of even-order distortion (Recio et al. 1997). This is consistent with the dearth of harmonic distortion in BM responses to tones at basal cochlear sites in chinchilla (Ruggero et al. 1997) and guinea pig (Cooper 1998) and with the similarity between BM h₁s and responses to clicks (Recio et al. 1998). These quasi-linear relations, which may seem paradoxical in the context of the prominent level-and frequency-dependent compressive nonlinearities of BM vibrations (Robles and Ruggero 2001), constitute a striking, albeit poorly understood, aspect of cochlear mechanics (see de Boer 1993; Recio et al. 1998; Zwislocki et al. 1997). Because the even-order nonlinearities revealed by the h₂s of high-CF ANFs do not arise in BM nonlinearities, they must arise at more central stages of cochlear signal encoding, including transduction currents and the generation of DC receptor potentials at the inner hair cells, as well as synaptic transmission, all processes characterized by asymmetrical input–output functions that produce rectification and/or other even-order nonlinearities (reviewed in Ruggero 1992).

Figure 12A compares the h₂-FSV of an ANF (CF: 9.5 kHz) and the h₁ for a BM site with similar CF (Recio et al. 1997). The same acoustic system and similar-level noise stimuli were used in recordings from 2 different chinchillas. The h₂-FSV from the ANF and the BM h₁ strongly resemble each other but the former is delayed relative to the latter.

FIG. 12 Comparison between the responses to noise of an ANF and a BM site with similar CFs. BM data are from cochlea L160, taken from Fig. 1 of Recio et al. (1997). BM site, located some 3.5 mm from the extreme basal end of the cochlea, and the ANF (from another (more ...)

Figure 12B presents the Fourier magnitudes of the BM h₁ of Fig. 12A and of other h₁s measured for higher-level stimuli at the same BM site. It is clear from Fig. 12B that BM and ANF responses of similar CF exhibited similar frequency tuning when stimulated by noise presented at low stimulus levels. For higher stimulus levels, the BM responses had increasingly broader tuning. A similar trend exists in responses to noise of ANFs [Figs. 11 and 12 of companion paper (Recio-Spinoso et al. 2005)], presumably as a result of the corresponding changes in BM vibrations.

In the companion paper, we presented near-CF group delays of chinchilla ANF responses to noise, computed from the phase-versus-frequency curves of h₂-FSVs [Fig. 15B of Recio-Spinoso et al. (2005)]. The trend line representing those data [derived from responses to low-level noise stimuli; see Fig. 16 of companion paper (Recio-Spinoso et al. 2005)] is reproduced in Fig. 13B (dashed line), which also includes a trend line (dotted line) for the near-CF group delays computed from phase-versus-frequency curves of chinchilla ANF responses to tones (Temchin and Ruggero 2001). The group delays for responses to tones (dotted line) and noise (continuous line) varied similarly as a function of CF but the former were somewhat shorter. The group-delay differences between the responses to tones and noise probably reflect the unequal stimulus levels (tones were presented at 60–70 dB, whereas noise stimuli were presented at near-threshold levels) and are consistent with the well-known inverse variation of group delays of ANF and BM vibrations as a function of stimulus intensity [e.g., Figs. 13 and 14 of the companion paper (Recio-Spinoso et al. 2005) and Fig. 1 of Ruggero (1994)].

The symbols in Fig. 13B depict near-CF group delays derived from cochlear mechanical responses at basal (Narayan and Ruggero 2000; Recio et al. 1997, 1998; Ruggero et al. 1997, 2000) and apical (Cooper 1997) sites of the chinchilla cochlea, with the upper and lower brackets indicating the range of group delays for near-threshold and intense stimuli, respectively. [Other phase-vs.-frequency curves for responses to tones at basal BM sites of the chinchilla cochlea have been published (Rhode and Recio 2000) but are not shown in Fig. 13B because they are available only for relatively intense stimuli (70 dB).] The synaptic and neural delays intervening between cochlear vibrations at the 3.5- and 14-mm sites of the chinchilla cochlea and the arrival of action potentials at a microelectrode near ANFs with appropriate CFs can be estimated by measuring the time differences between the trend lines for ANF responses and the symbols indicating BM vibrations. The solid black line, which is identical to the dashed curve but displaced downward by 1 ms, fits the BM data very well, indicating that on average action potentials arrive at the microelectrode in the auditory nerve about 1 ms after transduction in inner hair cells is initiated by deflection of their stereocilia. The 1-ms delay is identical to previous estimates for chinchilla ANFs from our laboratory (Ruggero and Rich 1987; Ruggero et al. 1996b) but somewhat longer than the synaptic delay (700–800 μs) estimated in the guinea pig cochlea (Palmer and Russell 1986).

Figure 13A compares the latencies of signal-front delays in ANFs and at the BM. The dashed line indicates the trend of the latencies of ANF responses to intense rarefaction clicks (from Fig. 10A). After subtracting 1 ms to compensate for synaptic/neural delays, the BM signal-front delays derived from ANF responses (continuous line) match almost perfectly the latencies directly measured in cochlear (BM and tectorial-membrane) responses to intense rarefaction clicks. This match supports the conclusion based on group delays (Fig. 13B) that a frequency-independent delay amounting to 1 ms intervenes between cochlea mechanical vibrations and the initiation of nerve impulses.

Finally, Fig. 14 compares the frequency glides observed in h₂-FSVs of ANFs and basilar membrane responses. The line is the trend line taken from Fig. 17 of the companion paper (Recio-Spinoso et al. 2005) and the symbols represent basilar membrane data obtained at several sites of the chinchilla cochlea. There is an excellent match between the neural and BM data in that both sets of data indicate high-to-low frequency glides for CFs <1 kHz and low-to-high glides for higher CFs. Interestingly, the neural data for CFs >4 kHz also support the contention, based on BM data, that the “relative amount of FM … decreases with increasing CF” at the base of the cochlea (Recio and Rhode 2000).

FIG. 14 Frequency glides in ANF and BM responses. Trend line is taken from Fig. 17C of the companion paper (Recio-Spinoso et al. 2005). A dimensionless glide slope was computed by dividing the slope expressed in kHz/ms by BF2 (i.e., kHz2); see Fig. 17 of companion (more ...)