Responses to diotic tone-in-noise stimuli in the inferior colliculus: stimulus envelope and neural fluctuation cues

Abstract

Human detection thresholds in tone-in-noise (TIN) paradigms cannot be explained by the prevalent power-spectrum model when stimulus energy is made less reliable, e.g., in roving-level or equal-energy paradigms. Envelope-related cues provide an alternative that is more robust across level. The TIN stimulus envelope is encoded by slow fluctuations in auditory-nerve (AN) responses - a temporal representation affected by inner-hair-cell (IHC) saturation and cochlear compression. Here, envelope-related fluctuations in AN responses were hypothesized to be reflected in responses of neurons in the inferior colliculus (IC), which have average discharge rates that are sensitive to amplitude-modulation (AM) depth and frequency. Responses to tones masked by narrowband gaussian noise (GN) and low-noise noise (LNN) were recorded in the IC of awake rabbits. Fluctuation amplitudes in the stimulus envelope and in model AN responses decrease for GN maskers and increase for LNN upon addition of tones near threshold. Response rates of IC neurons that are excited by AM were expected to be positively correlated with fluctuation amplitudes, whereas rates of neurons suppressed by AM were expected to be negatively correlated. Of neurons with measurable TIN-detection thresholds, most had the predicted changes in rate with increasing tone level for both GN and LNN maskers. Changes in rate with tone level were correlated with envelope sensitivity measured with two methods, including the maximum slopes of modulation transfer functions. IC rate-based thresholds were broadly consistent with published human and rabbit behavioral data. These results highlight the importance of midbrain sensitivity to envelope cues, as represented in peripheral neural fluctuations, for detection of signals in noise.

Keywords: Amplitude-modulation tuning; Low-noise noise; Masked detection.

Fig. 1.

Example stimulus waveforms and responses of AN and IC models illustrate the proposed envelope-based hypothesis for masked detection. Rows from top to bottom: stimulus waveform (with envelope highlighted in red), high-spontaneous-rate AN model responses (with neural fluctuation highlighted in red), band-enhanced (BE) IC model responses, and band-suppressed (BS) IC model responses. Columns from left to right are for 1/3-octave narrowband gaussian noise (GN), GN with tone, 1/3-octave low-noise noise (LNN), and LNN with tone. The tone frequency, center frequency of the noise maskers, and the model CFs were 1 kHz. Noise level was 65 dB SPL, and tone level was 0 dB SNR. Average rates of model responses are shown in the format of “masker-alone rate vs. masker-plus-tone rate” in the title. Model BE and BS neurons have decreasing or increasing rate upon addition of a tone to GN, opposite for LNN masker, consistent with envelope-based hypothesis. (Model AN responses based on Zilany et al., 2014; model IC responses based on Carney et al., 2015).

Fig. 2.

Distribution of MTFs across CF (in one-octave bins) for all units examined in this study. Gray shades from light to dark indicate units with band-enhanced (BE), band-suppressed (BS), hybrid, and all-pass (AP) MTF shapes. Three neurons with CF of 12.1k were included in the highest-CF bin. MTF types were represented across the range of CFs, except that hybrid MTFs were not observed at lower CFs. The largest numbers of units were in the frequency range for which rabbits have the most sensitive hearing (Heffner and Heffner, 2007).

Fig. 3.

MTFs and responses to tone-in-GN for two example neurons. (A) (B) MTFs of BE #1 (band-enhanced, BE) and BS #1 (band-suppressed, BS). The rate at each modulation frequency was an average of five repetitions. (C) (D) Average rate responses to tone-in-GN for BE #1 and BS #1, respectively. Colored lines with different symbols represent different overall noise levels. Filled symbols indicate supra-threshold SNRs based on ROC analyses of average rate. Tone frequency, near CF, is indicated in the title. BE #1 had decreasing rate as SNR increased, whereas BS #1 had increasing rate.

Fig. 4.

MTF and TIN responses of two example neurons. Left: MTFs. Right: Average rate in response to tone-in-GN (blue circles) and tone-in-LNN (red squares) at five overall noise levels. Red flat arrows and blue v-head arrows indicate thresholds for GN and LNN maskers, respectively; filled symbols indicate supra-threshold SNRs based on ROC analyses. Tone frequency (TF) for all TIN stimuli, which was near CF, is indicated at the right. Average rate for BE #2 decreased for tone-in-GN, except for a small increase for the 45-dB SPL noise; rate for tone-in-LNN increased. Average rate for BS #2 increased as SNR increased for the tone-in-GN and decreased for tone-in-LNN stimuli near threshold. LNN response curves change direction at high SNRs (see text).

Fig. 5.

MTF and TIN responses of five example neurons with complex TIN response patterns. Left: MTFs. Right: tone-in-GN (circle) or tone-in-LNN (square) responses at five noise levels. Same format as Fig. 4.

Fig. 6.

Rate-level functions (RLFs) of pure tone from response map (solid black), of GN responses (blue dot-dashed), and of LNN responses (red dashed). Titles show neuron numbers, from Figs. 3–5. Most neurons had non-monotonic RLFs, and noise RLFs often differed from tone RLFs (e.g., BS #1).

Fig. 7.

Proportion of neurons having decreasing (dark gray) or increasing (light gray) rate-changes among neurons with BE (left) and BS (right) MTF shapes in response to TIN with GN (upper plots) or LNN maskers (lower plots). Each bar shows results for one noise level. Expected rate-change directions are indicated by the black dots. The proportion was calculated based on the total number of units (shown at the top of each bar) that had a rate-based threshold at that noise level.

Fig. 8.

Correlation between TIN rate differences and maximum slope of the MTF. For each neuron, the TIN rate difference was calculated as the absolute maximum change in rate, across all SNRs, relative to the noise-alone condition.. A) GN masker. B) LNN masker. Each row represents one noise level. Symbols indicate MTF types (see legend). Correlation coefficient and the p value are shown in each plot; star indicates the correlation coefficient was significant after Bonferroni correction (α < 0.01). The TIN rate difference was significantly correlated with the maximum MTF slope for all noise levels and both masker types.

Fig. 9.

For each neuron, the TIN rate difference for the GN masker (A) and the LNN masker (B) at each noise level (rows) is plotted versus the rate difference between GN-alone and LNN-alone at each noise level. Symbols indicate MTF types (see legend). Correlation coefficients and p values are shown in each plot; stars indicate significant correlation coefficients after Bonferroni correction (α < 0.025). TIN rate difference was negatively correlated with the rate difference between the two GN maskers, and positively correlated for LNN maskers, as expected.

Fig. 10.

Correlation between TIN rate differences for GN and LNN maskers for single neurons. MTF types are indicated by different symbols (see legend). Correlation coefficients and p values are shown in each plot. The maximum rate differences between TIN responses to the two noise maskers were significantly positively correlated at noise levels of 35 to 45 dB SPL, not significantly correlated at 55 to 65 dB SPL, and significantly negatively correlated at 75 dB SPL.

Fig. 11.

Rate-based thresholds for all units at each noise level with GN (A) and LNN (B) maskers. Thresholds of the most sensitive neurons across frequency were similar to human behavioral data and to rabbit behavioral data, which is only available at 500 Hz. (C) Rate-based TIN threshold differences between GN masker and LNN masker conditions (open squares). At each noise level, only neurons with interpolated thresholds for both maskers were included. More neurons had higher threshold for the GN masker than for the LNN masker, consistent with human behavioral data (Goupell, 2012; Kohlrausch et al., 1997).