Noise-induced enhancement of envelope following responses in normal-hearing adults

Abstract

Measures of signal-in-noise neural encoding may improve understanding of the hearing-in-noise difficulties experienced by many individuals in everyday life. Usually noise results in weaker envelope following responses (EFRs); however, some studies demonstrate EFR enhancements. This experiment tested whether noise-induced enhancements in EFRs are demonstrated with simple 500- and 1000-Hz pure tones amplitude modulated at 110 Hz. Most of the 12 young normal-hearing participants demonstrated enhanced encoding of the 110-Hz fundamental in a noise background compared to quiet; in contrast, responses at the harmonics were decreased in noise relative to quiet conditions. Possible mechanisms of such an enhancement are discussed.

1. Introduction

Measures of neural encoding can be employed as tools to improve our understanding of perception difficulties for signals presented in background noise; for example, the steady-state or following responses of the human auditory system provide information about auditory coding of ongoing temporal and spectral acoustic cues important for speech understanding in noise. These auditory evoked potentials are thought to be a measure of phase locking, providing information about the neural coding primarily in the brainstem. Several studies have demonstrated relationships between following responses and speech perception in noise (Anderson et al., 2011; Bidelman et al., 2014; Parbery-Clark et al., 2011; Hornickel et al., 2009; Song et al., 2011), suggesting the potential to use these measures to improve our understanding of perception-in-noise difficulties.

The envelope following response (EFR), when elicited with stimuli presented in background noise, can offer a means to characterize suprathreshold encoding in the auditory system, and perhaps, to explain the variability observed across individuals for perception-in-noise tasks. Most often, the presence of background noise decreases perception and weakens the neural coding of target signals; however, there are some circumstances in which the presence of background noise has been associated with improvement in behavioral thresholds (Ries, 2007) and increases in the amplitude of cortical auditory evoked responses (Alain et al., 2009; Papesh et al., 2015). This improvement may be a manifestation of stochastic resonance-like properties of the auditory nervous system. Depending on the stimuli, response metric, and participant group, following responses can demonstrate enhancements to the magnitude of the neural response at the fundamental frequency (Leigh-Paffenroth and Murnane, 2011; Osman et al., 2016; Prévost et al., 2013). There is also evidence of minimal magnitude change or decrements in magnitude with background noise; generally, encoding of the fundamental remains robust in background noise until signal-to-noise ratios (SNRs) are poor, resulting in magnitude decrements (Li and Jeng, 2011; Anderson et al., 2013; Song et al., 2011; Osman et al., 2016; Laroche et al., 2013).

In the published instances where fundamental frequency enhancement in background noise has been observed, complex tones or synthesized vowel stimuli were used to elicit the EFR response. This raises the possibility that the enhancement is due to interactions among stimulus components in the ear (e.g., canceling/adding) that are then reflected at the level of the scalp (Leigh-Paffenroth and Murnane, 2011). In this study, we used simple amplitude-modulated tones to avoid possible interactions among stimulus components. Additionally, because typically only group data have been reported, little is understood about the variability of noise-induced enhancement effects across individuals. To address this, we report individual EFR enhancement effects in response to sinusoidally amplitude modulated tones.

2. Materials and methods

2.1. Subjects

Twelve young normal-hearing participants (9 male, 3 female), ranging in age from 19 to 36 yr, were tested in this study. Participants reported good general health and had normal hearing (i.e., 25 dB hearing level or better) in both ears. All participants passed the Mini-Mental State Examination (Folstein et al., 1975) with a score of 29 or better. The research protocol was approved by the Veterans Affairs Portland Health Care System Institutional Review Board and participants provided written informed consent.

2.2. Stimuli

Stimuli were three separate amplitude modulated pure tones that had a carrier frequency of 500, 1000, or 4000 Hz. The amplitude modulation frequency was always 110 Hz at a depth of 100%. Stimuli had durations of 200 ms including 20-ms cosine squared rise/fall ramps. Three conditions were presented: (1) 80 dB sound pressure level (SPL) with no background noise (s80n00), (2) 50 dB SPL with no background noise (s50n00), and (3) 80 dB SPL with 60 dB SPL of a continuous speech-spectrum background noise (s80n60) that has been used previously (Billings et al., 2015). In addition, an artifact run was also completed for each subject using the s80n00 condition for a randomized carrier frequency. Because time constraints prevented testing of the s80n60 condition for the 4000-Hz carrier, the 12 participants were tested for a total of nine conditions, which were presented randomly for each participant. As the focus of this paper is the enhancement resulting from the presence of background noise (i.e., increases in response strength for s80n60 relative to s80n00), data for the 4000-Hz carrier are not presented.

2.3. Electrophysiology

Tin electrodes placed at Cz, Fz, C7, and M1 were active with a reference electrode at M2 (right mastoid) and a ground electrode at Fpz. Neuroscan SynampsRT and Scan 4.5 (Compumedics, Neuroscan, Charlotte, NC) were used for data collection with an acquisition sampling rate of 20 000 Hz and a bandpass filter setting of 100–3000 Hz at 6 dB/octave. For each condition, the stimulus was presented 3000 times with stimulus polarity alternating for each sweep, resulting in 2 bins of 1500 sweeps corresponding to non-inverted and inverted stimulus polarities, respectively. Alternating polarities were added to highlight the response to stimulus envelope and reduce artifacts such as the cochlear microphonic (Aiken and Picton, 2008). Three inter-stimulus intervals (offset to onset) were randomly used within each block with equal proportion of usage: 66.67, 83.33, and 100 ms. Stimuli were presented to the right ear through an electrostatically shielded Etymotic ER3 insert transducer with shielded cables.

During testing participants were encouraged to relax and sleep. As mentioned previously, individuals completed an additional artifact condition using the s80n00 stimulus in which the transducer tip was removed from the test ear, and both ears were plugged with generic foam earplugs. Artifact test conditions revealed no obvious stimulus artifact present for EFRs. There were cases in various conditions where the 60-Hz artifact was present. We elected to not use a 60-Hz notch filter and instead excluded multiples of the powerline from the spectral analysis so as to avoid the possibility of including powerline artifacts in test results.

2.4. Data processing and analysis

Continuous electroencephalography was epoched in Neuroscan Edit 4.5 and then exported to matlab, where each epoch was bandpass filtered using a zero-phase 800th order finite impulse response filter from 80 to 4140 Hz to ensure no residual activity outside these frequencies was present. The pre- and post-stimulus interstimulus intervals were included in each epoch for filtering steps to prevent filtering edge effects (e.g., delay in the neural response). Of particular importance was ensuring that any energy at 60 Hz was completely filtered out. Sweeps were then processed to reject the noisiest epochs using an artifact rejection criterion of 60 μV. Using this criterion, at least 1000 epochs were accepted of each polarity in each condition.

Two different measures of EFR strength were used in this study: (1) temporal coherence between the acoustic stimulus and neural response was assessed using a stimulus-to-response correlation coefficient (SRCC), and (2) response strength was assessed using a SNR. The SRCC is derived from the normalized cross-correlation function between the stimulus envelope and the response, both of which were 200 ms. The cross-correlation function between two discrete real-valued signals sharing a common sampling rate involves computing the dot-product (pointwise multiplication followed by summation across points) as a function of the lag—that is, the degree of displacement in the time-alignment of one signal relative to the other. This function can be normalized by dividing by the geometric mean of the sum-of-squares of each signal. This makes the value at each point equivalent to a Pearson correlation coefficient. The SRCC was defined as the peak absolute value of the normalized cross-correlation function within the allowed lag range of 2–22 ms. This range was designed to encompass all physiologically reasonable neural latencies (0–20 ms, based on King et al., 2016), with 2 ms added to account for insert-earphone tube delay. Individual lags were estimated using methods previously established (Billings et al., 2019). The SNR magnitude is expressed as the ratio of signal and noise voltages and was calculated in decibels (20 times the base-10 logarithm) from the discrete Fourier transform of each neural response in the time domain. Frequencies of interest included the fundamental frequency (110 Hz) and its harmonics (e.g., 220, 330, 440, and 550 Hz). The SNR in dB for each frequency of interest was calculated using the average of ten frequency bins surrounding the frequency of interest (five bins from each side) after skipping the neighboring frequency bin on either side of the frequency of interest to avoid spectral leakage. If one of the neighboring bins was a harmonic of the 60-Hz power line frequency, this bin was not included and an additional adjacent bin was added to compute the noise floor.

3. Results

Figure 1 shows the grand averaged EFR temporal waveforms (left) and discrete Fourier transform spectra (right) for the 500- (top) and 1000-Hz (bottom) carriers as a function of condition. Figures and analyses include all responses. There are minimal changes to the temporal waveforms as a function of carrier frequency; however, differences among the conditions (s80n00, s50n00, and s80n60) are more obvious. Spectral energy at the fundamental modulation frequency of 110 Hz (Fig. 1, right) is present for all conditions, whereas responses at subsequent harmonics (220, 330, 440, and 550 Hz) are smaller for the s80n60 and s50n00 conditions relative to the quiet condition (s80n00). In addition, there appears to be an increase in the response at the fundamental and a decrease of the response at the harmonics for the stimulus presented in noise (s80n60) relative to quiet (s80n00) for both carrier frequencies presented (see green vs black lines in both time and frequency domains). Table 1 shows the posterior 90% Bayesian confidence intervals and probability of condition differences for SNR and SRCC outcome measures at both the 500- and 1000-Hz carriers. For SNR at the 110-Hz fundamental and for SRCC, neural responses are most robust for s80n60 followed by s80n00. In contrast, the SNR at the 220-Hz harmonic shows a different pattern: s80n00 is the most robust followed by s80n60. In all cases, the s50n00 condition demonstrates the weakest responses. Comparing the noise condition with the no-noise condition reveals a dissociation between the response at the fundamental and the response at the harmonics (i.e., an increased response in noise at the 110-Hz fundamental contrasts with the reductions in strength at the 220-Hz harmonic).

Fig. 1.

(Color online) Grand averaged (n = 12) time waveforms (left) and spectral waveforms (right). Results for the two carriers (500, 1000 Hz) are shown for the three stimulus conditions of 80-dB signal in quiet (s80n00), 80-dB signal with 60 dB of background noise (s80n60), and 50-dB signal in quiet (s50n00). The different stimulus conditions evoke distinct following responses especially at the 110-Hz fundamental. At the second harmonic (220 Hz), only the s80n00 condition appears to have a response above the noise floor. Overall, brainstem encoding of the fundamental is more robust in the presence of background noise when compared to the other conditions.

Table 1.

Posterior estimates of condition differences. Note: LCL/UCL, lower or upper 90% Bayesian confidence limits.

		Contrasts
		s80n60–s80n00		s80n60–s50n00		s80n00–s50n00
SRCC		90% Interval (LCL/UCL)	Probability s80n60 >s80n00	90% Interval (LCL/UCL)	Probability s80n60 >s50n00	90% Interval (LCL/UCL)	Probability s80n00 >s50n00
	500-Hz Carrier	0.0972/0.2321	100%	0.2006/0.3395	100%	0.0631/0.1550	100%
	1000-Hz Carrier	0.0415/0.1608	100%	0.0951/0.2208	100%	0.0094/0.0994	98%
SNR
	500-Hz Carrier
	110-Hz Fundamental	6.6340/19.602	100%	24.1918/37.4214	100%	11.2317/24.1390	100%
	220-Hz Harmonic	−20.937/-8.050	0%	−0.6848/11.8681	94%	13.9401/26.5128	100%
	1000-Hz Carrier
	110-Hz Fundamental	0.5367/13.5156	96%	9.9629/23.6971	100%	3.4712/16.4668	100%
	220-Hz Harmonic	−10.9205/0.3747	4%	1.8357/14.3571	99%	7.5523/18.9292	100%

The EFR across individuals is shown in Fig. 2. While the general trends seen in Fig. 1 are evident, Fig. 2 demonstrates the wide range of recording quality across participants. For example, participants 3 and 5 are illustrative of more noisy recordings, making identification of the fundamental and harmonics difficult. Across participants, the response at the 110-Hz fundamental is robust at both carrier frequencies for s80n00 and s80n60 but is less reliably detected for s50n00. The 220-Hz spectral response is diminished or absent relative to the fundamental across all participants and conditions, and the decrement is more noticeable across individuals at the higher carrier frequency of 1000 Hz when compared to the 500-Hz carrier responses.

Fig. 2.

(Color online) Spectral waveforms for all conditions and individuals tested. A broad range of response quality is exhibited across all participants with participants 3 and 5 showing a minimal/noisy response. The 110-Hz fundamental response is present in most participants for both carriers while the 220-Hz harmonic response is less robust for the 1000-Hz carrier. The fundamental response in the temporal waveform is enhanced in the presence of noise for most individuals tested.

The comparison between the s80n00 condition and the s80n60 condition was of particular interest in this study given the mixed findings of EFR enhancements in background noise relative to quiet in the literature. Figure 3 illustrates the presence of enhancements or decrements when comparing s80n00 and s80n60 across individuals (individuals are shown in thin lines and the mean is the thick black line). Enhancements for s80n60 relative to s80n00 are generally seen for both SRCC and 110-Hz SNR measures, whereas decrements are generally seen for the 220-Hz SNR measure. For the 500-Hz carrier stimulus, all 12 participants demonstrate enhancements for the SRCC and 110-Hz SNR measures. The mean SRCC enhancement is 0.18 and ranges between 0.03 and 0.33, and the mean 110-Hz SNR enhancement is 18 dB with a range of 3 to 37 dB. Decrements for s80n60 relative to s80n00 are evident for 10 of 12 participants for the 220-Hz SNR measurement with a mean decrement of 17 dB. For the 1000-Hz carrier stimulus, 9 of 12 participants demonstrate SRCC enhancements; 8 of 12 participants demonstrate enhancements for the 110-Hz SNR measure. The mean SRCC enhancement is 0.10, and the mean 110-Hz SNR enhancement is 8 dB. Decrements are exhibited for 10 of 12 participants for the 220-Hz SNR measurement with a mean decrement of 7 dB.

Fig. 3.

(Color online) Individual SNR and SRCC enhancement effects (i.e., a comparison between s80n60 and s80n00 conditions). SNR and SRCC values for each participant are shown in the thin lines, and mean values are shown with the thick black line. The SRCC (right column) and 110-Hz SNR (left column) generally demonstrate an enhancement when noise is included in the stimulus while the 220-Hz SNR (middle column) generally shows a diminished response. The degree of enhancement varies widely across participants and appears less consistent for the 1000-Hz carrier as compared to the 500-Hz carrier.

4. General discussion and conclusion

Our interest in the EFR stems from its potential use as a clinical tool (for a review see Korczak et al., 2012) to quantify variability present in the brainstem-level neural coding across individuals who vary in age, hearing loss, or exposure to noise (Anderson et al., 2013; Bharadwaj et al., 2015; Clinard and Tremblay, 2013; Vander Werff and Burns, 2011; Yamada et al., 1978). In particular, correlations between the EFR and measures of speech perception in noise suggest that auditory brainstem measures can inform our understanding of the neural encoding of signals in noise (Bidelman et al., 2014; Hornickel et al., 2009; Song et al., 2011; Anderson et al., 2011). The aim of this study was to clarify the existing literature with regards to enhancements in the EFR. We investigated the basic effects of carrier frequency and stimulus level on the EFR with emphasis on how background noise affects the response.

As expected, EFRs to the two carriers used in this study (500 and 1000 Hz) were quite similar. Picton and colleagues (2009) have demonstrated stable EFR amplitudes across a range of carriers (500–4000 Hz). In addition, Picton et al. (2005) and others have demonstrated a clear, systematic effect of stimulus level, with increases in level resulting in increases in EFR response amplitude—an effect confirmed in these data.

Noise-induced enhancement of the EFR at the fundamental, but not the harmonics, was present for both carrier frequencies and for most participants. The enhancement we observed in response to amplitude modulated single pure tones presented in noise is novel and extends the enhancement literature which has reported only occasional enhancements in response to more complex stimuli such as tonal complexes or synthesized vowels (Leigh-Paffenroth and Murnane, 2011; Osman et al., 2016; Prévost et al., 2013). Most notably, enhancements of the SNR and the SRCC measures for the 110-Hz fundamental in background noise, relative to the quiet condition, were found for all participants with the 500-Hz carrier stimulus, and for most of the participants with the 1000-Hz carrier (8/12 participants for SNR and 9/12 for SRCC). It is unclear why a third of subjects lacked enhancement for the 1000-Hz carrier stimulus, except perhaps that overall differences in physiological noise may have led to the response being buried in the noise floor for those participants.

It is noteworthy that previous studies have reported fundamental frequency enhancement in response to complex stimuli, either multiple tones (Leigh-Paffenroth and Murnane, 2011) or synthetic vowels (Osman et al., 2016; Prévost et al., 2013), but not for simple stimuli. Nevertheless, in the current study, when present, enhancements were observed in response to simple amplitude modulated pure tones (in which no interaction of signal components is possible) in a noise background. The increase in SNR and temporal coherence (SRCC) in background noise found here and in other studies are generally restricted to the fundamental frequency; whereas decrements generally occur at harmonic frequencies. Still, the current individual data exhibit several exceptions for the 1000-Hz carrier: four individuals displayed decrements at the fundamental, and two individuals displayed increments at the harmonic for the SNR; similarly, three individuals had temporal coherence decrements. It appears that enhancements are more universally observed at lower carrier frequencies, which may provide a hint to the underlying mechanism involved.

Various mechanisms have been proposed to underlie neural enhancement: Prévost et al. (2013) suggested that synchrony capture may play a role in this effect. Synchrony capture refers to the enhanced neural activity close to a target frequency, such as the fundamental, and suppression of neural activity outside that target area (Young and Sachs, 1979). Previous studies show that neural responses to harmonics of the fundamental, or even the second formant, decrease in amplitude in the presence of background noise relative to quiet conditions. Our data support the idea that there is dissociation between the EFR to the fundamental and its harmonics in the presence of background noise. Synchrony capture within subcortical auditory nuclei may help extract the pitch of the low frequency formants by enhancing neural encoding of the strongest harmonic present in each formant region. Further studies are needed to better isolate the mechanisms involved and to determine how these EFR characteristics relate to speech understanding in noise.