Speech masking release in Hybrid cochlear implant users: Roles of spectral and temporal cues in electric-acoustic hearing

Abstract

When compared with cochlear implant (CI) users utilizing electric-only (E-Only) stimulation, CI users utilizing electric-acoustic stimulation (EAS) in the implanted ear show improved speech recognition in modulated noise relative to steady-state noise (i.e., speech masking release). It has been hypothesized, but not shown, that masking release is attributed to spectral resolution and temporal fine structure (TFS) provided by acoustic hearing. To address this question, speech masking release, spectral ripple density discrimination thresholds, and fundamental frequency difference limens (f₀DLs) were evaluated in the acoustic-only (A-Only), E-Only, and EAS listening modes in EAS CI users. The spectral ripple and f₀DL tasks are thought to reflect access to spectral and TFS cues, which could impact speech masking release. Performance in all three measures was poorest when EAS CI users were tested using the E-Only listening mode, with significant improvements in A-Only and EAS listening modes. f₀DLs, but not spectral ripple density discrimination thresholds, significantly correlated with speech masking release when assessed in the EAS listening mode. Additionally, speech masking release correlated with AzBio sentence recognition in noise. The correlation between speech masking release and f₀DLs likely indicates that TFS cues provided by residual hearing were used to obtain speech masking release, which aided sentence recognition in noise.

I. INTRODUCTION

Modifications in cochlear implant (CI) electrode array design, coupled with soft surgical techniques, have allowed CI users to retain acoustic hearing in the implanted ear post-surgery. As a result, these electric-acoustic stimulation (EAS) CI systems provide electrical stimulation of high-frequency sounds and acoustic amplification of low-frequency sounds. Preservation of acoustic hearing can maintain access to spectral/temporal cues that patients previously used preoperatively (Gifford et al., 2008). Adding acoustic amplification to electric hearing increases word recognition in quiet compared to electric-only (E-Only) hearing. The acoustic benefit in noise, however, is greater than in quiet (Gantz and Turner, 2003; Gantz et al., 2005; Gifford et al., 2008; Gstoettner et al., 2008; Helbig et al., 2011; Kim et al., 2017; Turner et al., 2004; Turner et al., 2008; Turner et al., 2010). The amount of preserved acoustic hearing has been correlated with EAS benefit in noise (Turner et al., 2008; Gifford et al., 2013). Preserved acoustic hearing has also been hypothesized to lead to speech masking release, as seen in a small cohort of EAS CI users (n = 3; Turner et al., 2004), but this hypothesis has not been evaluated in a large subject population. The current study aims to ascertain the spectral/temporal cues provided by residual hearing that lead to speech masking release in a large cohort of EAS CI users.

Speech masking release is the improvement in speech recognition in temporally fluctuating noise compared to steady-state noise. Glimpses of speech present within the valleys of fluctuating noise (e.g., multi-talker babble, amplitude-modulated noise) allow one to “listen in the dips” and improve speech recognition. Normal hearing (NH) listeners demonstrate robust masking release (Carhart et al., 1969; Festen and Plomp, 1990; Bacon et al., 1998; Peters et al., 1998; Nelson et al., 2003; Nelson and Jin, 2004; Stickney et al., 2004; Goldsworthy et al., 2013; Goldsworthy, 2015), while hearing-impaired (HI) listeners show less masking release even when audibility is accounted for (Festen and Plomp, 1990; Takahashi and Bacon, 1992; Eisenberg et al., 1995; Bacon et al., 1998; Peters et al., 1998; Dubno et al., 2002; Turner, 2006; but see Gustafsson and Arlinger, 1994). Minimal masking release is observed in CI users and NH listeners tested using vocoded speech, likely due to degraded spectral and/or temporal cues provided by CI signal processing algorithms (Nelson et al., 2003; Nelson and Jin, 2004; Stickney et al., 2004; Turner et al., 2004; Fu and Nogaki, 2005; Jin et al., 2013; Goldsworthy et al., 2013; Goldsworthy, 2015; Croghan and Smith, 2018). A preliminary study of three CI users implanted with the Cochlear Ltd. S8 Hybrid hearing preservation electrode array (six-electrode array; 10-mm insertion depth; Sydney, Australia) showed evidence of masking release using EAS and suggested that preserving low-frequency acoustic hearing led to masking release (Turner et al., 2004). However, it is not clear what specific spectral/temporal cues provided by residual acoustic hearing give rise to speech masking release in EAS CI users. In addition, the S8 Hybrid array was implanted under investigational trials and is not available for clinical use; thus, the preliminary findings of EAS masking release need to be confirmed in a sufficiently large group of patients using a current generation, clinically available, hearing preservation electrode array.

There is a hypothesized role of spectral resolution in masking release given that NH groups listening to vocoded speech show greater masking release with an increasing number of analysis bands (Qin and Oxenham, 2003; Nelson et al., 2003; Nelson and Jin, 2004; Fu and Nogaki, 2005) and steeper carrier band filter slopes (simulating less spectral smearing; Fu and Nogaki, 2005). The increased spectral resolution offered by additional analysis bands may allow one to resolve spectral information present in speech within the valleys of the masker, leading to masking release.

Additionally, spectral resolution and temporal fine structure (TFS) are thought to affect pitch perception and impact speech understanding in noise (Moore, 2008; Oxenham, 2008, for reviews). The reduced (and sometimes absent) TFS cues provided by CI signal processing algorithms force CI listeners to obtain weak pitch information from temporal envelope cues and limited spectral/spatial cues, subsequently affecting fundamental frequency (f₀) discrimination (Qin and Oxenham, 2005; Schvartz-Leyzac and Chatterjee, 2015) and stream segregation (Hong and Turner, 2006; Tejani et al., 2017). Improved f₀ perception provided by low-frequency acoustic hearing can allow for segregation of the target from the masker, leading to improved speech understanding in noise for CI users (Qin and Oxenham, 2006; Brown and Bacon, 2010; Zhang et al., 2010) and speech masking release in EAS listening (Turner et al., 2004; Hopkins and Moore, 2009).

The EAS benefits for speech recognition have been hypothesized to arise from preserved acoustic hearing providing access to spectral and temporal cues that are not sufficiently coded electrically. However, psychophysical measures of spectral and temporal processing have not correlated with EAS benefits (Gifford et al., 2008; Gifford et al., 2010; Gifford and Dorman, 2012). The lack of correlation may be due to the definition of EAS benefit. The aforementioned studies by Gifford and colleagues, as well as others (e.g., Usami et al., 2014; Kim et al., 2017; Pillsbury et al., 2018), quantified EAS benefit as the improvement in speech recognition when using EAS hearing relative to using E-Only hearing. This operational definition relies on artificially constructed test conditions. Testing in the E-Only mode requires the EAS CI user to listen to a high-pass filtered speech signal, while testing in the acoustic-only (A-Only) mode reflects understanding of low-pass filtered speech. However, EAS CI users are acclimated to combined electric and acoustic stimulation, which allows access to a broadband speech signal. Consequently, E-Only scores for EAS CI users may be lower than E-Only scores of patients with conventional implants due to the limited bandwidth and lack of practice in that listening mode (e.g., Dorman et al., 2009). This definition of EAS benefit may not be optimal, especially when trying to correlate EAS benefit with spectral and temporal processing capabilities of residual acoustic hearing.

The current study addressed three goals. The first goal was to evaluate EAS benefit via a speech masking release paradigm. Evaluation of masking release is an ecologically valid metric because listeners often attend to target speech in temporally fluctuating competing speech. The current study utilized a closed-set spondee recognition paradigm as this allowed for speech testing at a negative signal-to-noise ratio (SNR) without floor effects (Miller et al., 1951) and promoted greater masking release compared to open-set testing (Buss et al., 2009). The negative SNR was necessary because masking release is greater for negative SNRs, implying that listening in the dips is necessary for more difficult noisy backgrounds (Takahashi and Bacon, 1992; Dubno et al., 2002; Nelson et al., 2003; Nelson and Jin, 2004; Stickney et al., 2004; Bernstein, 2011, for review). An adaptive procedure was not used because the resulting speech reception thresholds (SRTs) vary amongst individuals, which consequently affects masking release (Bernstein, 2011, for review). To support our claim that speech masking release is ecologically valid, masking release was also correlated with clinical speech recognition in noise performance.

The second goal was to evaluate spectral and temporal processing that could contribute to masking release. Spectral ripple density discrimination thresholds and fundamental frequency difference limens (f₀DLs) were measured. Spectral ripple stimuli are broadband noise with spectral peaks linearly or logarithmically spaced on the frequency scale. Listeners compare two ripple stimuli that are identical except that the location of the spectral peaks and valleys are out of phase. As the spacing between ripples is reduced (higher ripple density), it becomes harder to discriminate between the standard and out-of-phase stimulus. Spectral ripple density discrimination thresholds correlate with closed-set consonant and vowel recognition in quiet across NH, HI, and CI groups (Henry and Turner, 2003; Henry et al., 2005; Anderson et al., 2011) and with speech recognition in noise using closed-set spondee and open-set sentences for standard CI listeners (Won et al., 2007; Jeon et al., 2015; but see Anderson et al., 2011, who reported trends approaching significance). Trends approaching significance also have been reported in HI listeners for open-set speech recognition in noise (Davies-Venn et al., 2015). Additionally, ripple discrimination thresholds correlate with spatial tuning curves in CI users (Anderson et al., 2011), providing evidence that the ripple task at least partially probes underlying spectral/spatial resolution.

The f₀DL task reflects one's sensitivity to frequency changes between stimuli. f₀DLs correlate with SRTs in noise across NH and HI listeners (Glasberg and Moore, 1989) and with identification of simultaneously presented vowels (Summers and Leek, 1998; but see Vongpaisal and Pichora-Fuller, 2007, who reported similar but non-significant correlations). More recently, Goldsworthy and colleagues correlated f₀DLs with speech perception in noise and speech masking release for consonants and vowels in standard CI users, although the use of multiple comparisons may have precluded any significant findings (Goldsworthy et al., 2013; Goldsworthy, 2015).

The last goal was to conduct all measures in three listening modes (A-Only, E-Only, and EAS) to understand the contributions of acoustic and electric hearing to masking release, spectral ripple density discrimination performance, and f₀ sensitivity. With respect to speech masking release, this design admittedly does not completely resolve the drawbacks of A-Only and E-Only speech testing as previously discussed (artificial listening conditions, limited bandwidth). However, the use of simple spondees and closed-set testing arguably reduced the listening and cognitive effort needed to recognize the spondees (e.g., Miller et al., 1951) as opposed to open-set testing done in clinical environments. This allows subjects to allocate cognitive resources toward dip-listening. With respect to spectral/temporal processing, unlike standard CI users, EAS CI users show better sentence recognition as f₀ separation between target and masker increases, indicating that they can use acoustic hearing to resolve f₀ differences (Auinger et al., 2017). Similarly, EAS CI users outperform standard CI users on spectral ripple density discrimination tasks. This improvement in ripple density discrimination remained when EAS CI users were tested using just their acoustic hearing alone, implying that much of the benefit arose from access to acoustic hearing (Golub et al., 2012).

Given our experimental goals, we hypothesized:

• (1)Speech masking release, spectral ripple density discrimination, and f₀DL performance in the A-Only and EAS modes will be similar to one another and will also be better than E-Only mode.
• (2)Spectral ripple density discrimination thresholds will directly correlate and f₀DLs will inversely correlate with masking release.
• (3)Greater masking release, higher (better) spectral ripple density discrimination thresholds, and lower (better) f₀DLs will be seen in EAS CI users with more residual acoustic hearing.
• (4)Speech masking release will directly correlate with clinical evaluation of speech recognition in noise (AzBio sentence recognition in noise).

II. METHODS

A. General methods

This study was approved by the University of Iowa Institutional Review Board. Subjects signed a consent form and provided consent for retrospective and prospective data collection.

Experimental interfaces and stimuli were developed using matlab 2014 b (The MathWorks, Natick, MA) and incorporated the “adaptive track engine” function (matlab File Exchange; Brimijoin, 2013). Stimuli were presented in a sound-treated booth via a loudspeaker at 65 dBA, representing a normal conversational level.

Subjects were tested in three listening conditions—EAS, A-Only, and E-Only—using their clinically assigned speech processor with no alterations made to the programming. Standard earplugs and an earmuff placed over the earplugs were used to attenuate acoustic stimuli in the non-implanted ear (and the implanted ear in the case of E-Only testing). Plugging and muffing resulted in 25–40 dB of attenuation for frequencies between 125 and 1000 Hz and 25 dB of attenuation for SRTs in the sound field. For E-Only testing, the acoustic component was removed from the ear canal. For A-Only testing, the transmitting coil was removed from the head to eliminate electrical stimulation. Note that the “auto processor off” feature was temporarily disabled in the speech processor MAP since this feature shuts off the sound processor when the coil is off the patient's head for an extended period, as in the case of A-Only testing.

Subjects underwent speech masking release, spectral ripple density discrimination, and f₀DL testing. The order of testing and the order of listening conditions (A-Only, E-Only, and EAS) within each test were randomized.

B. Subjects

All subjects (n = 22) were adults aged 18 years or older with at least 1 year of EAS CI use. All except subject L82R were patients of the University of Iowa Hospital and Clinics (UIHC) CI program. Subject L82R was implanted and receives clinical care at an outside clinic. Figure 1 shows postoperative unaided audiograms at the time of testing. Supplemental Table 1 provides frequency-specific audiometric thresholds for all subjects.¹ Table I provides participant demographics, speech processor programming details, and pre- and postoperative Consonant-Vowel-Nucleus-Consonant (CNC) and AzBio speech recognition scores derived from clinical records. CNC word recognition (Peterson and Lehiste, 1962) was assessed at 60 dBA in quiet. AzBio sentence recognition (Spahr et al., 2012) was assessed at 60 dBA in multi-talker babble at +5 dB SNR. Preoperative scores were assessed using a hearing aid in the ear to be implanted, while postoperative scores were obtained in the implanted ear at the time of testing with the contralateral ear plugged/muffed. The ipsilateral ear was plugged as well in the case of E-Only testing.

FIG. 1.

(Color online) Postoperative audiograms for the implanted ear and contralateral ear. Open symbols on the right plot represent mean ±1 standard deviation (SD). NR indicates no response. Vertical lines on the right plot indicate the average high-pass cutoff frequency for electrical stimulation and the average low-pass cutoff frequency for acoustic stimulation.

TABLE I.

Participants' audiological history. PTA, pure tone average of 125–500 Hz; E-Only, electric-only hearing; EAS, electric and acoustic hearing in the implanted ear. Electrodes 1–4 are deactivated by default.

Subject	Age (years)	Device use (months)	Implanted ear PTA (dB HL)	Contralateral ear PTA (dB HL)	Acoustic cutoff (Hz)	Electric cutoff (Hz)	Deactivated electrodes	Maxima	Channel stimulation rate (Hz)	CNC preoperative (%)	CNC E-Only (%)	CNC EAS (%)	AzBio preoperative (%)	AzBio E-Only (%)	AzBio EAS (%)
L36R	67	21	17	22	N/A	683		8	900	14	36	72	6		64
L50R	66	12	32	32	1000	563	5–8	5	250	22	30	40	11	21	40
L25L	21	36	72	42	500	438		7	900	46	67	70	44	15	20
L59R	75	12	70	68	500	313		8	900	12	58	73	0	0	1
L13L	51	60	63	63	500	688		8	900	2	62	81	1	16	50
L74L	73	25	43	20	1500	1000	5	8	900	32	62	69		23	10
L37R	40	24	43	40	500	563		8	900	35	80	96	10	24	64
L33L	64	31	65	37	250	313	5	8	500	12	33	42	11		12
L44R	54	25	38	38	750	563	5, 18	8	900	16	44	72	30	28	66
L6R	70	76	70	68	500	563		8	900	3	34	73	1		54
L2L	70	95	60	52	500	438		10	1200	16	74	84	15	42	50
L3R	48	93	55	47	500	813	13	8	900	16	70	87	12	49	65
L21L	61	48	48	50	750	813		8	900	13	61	84	1	23	71
L42R	77	24	48	48	1000	813		8	900	17	39	57	16	1	10
L48R	72	24	40	33	1000	813	5, 6, 12, 13	8	900	12	10	35	0	5	9
L82R	34	27	58	12	750a	438	5, 6, 7	8	900	48		76	40		52
L18R	40	49	73	28	500	438	5	8	900	35	70	83	33	18	71
L67R	64	12	62	53	750	563		8	900	0		70	0		31
L43R	78	23	48	42	750	813		8	900	4	32	83		4	46
L19L	53	47	75	37	500	438		8	900	46	83	89	37	53	80
L70L	62	13	55	47	750	563		8	900	25	49	81		5	56
L1R	31	106	83	62	250	188		7	900	24	56	72	6	17	31

^aSubject's acoustic component under targets. See the text for details.

All participants utilized the Cochlear Ltd. L24 Hybrid electrode array, a 22-electrode array (16 mm insertion depth), which is approved by the U.S. Food and Drug Administration (FDA) for EAS purposes. Our participants were reasonably representative of the general L24 Hybrid CI patient population with an average age of 58 ± 16 years and average EAS speech perception scores of 72 ± 16% (CNC in quiet) and 43 ± 24% (AzBio in +5 dB SNR) at the time of testing. For comparison, the average age was 64.1 ± 14.7 years for the 50 patients in the U.S. L24 Hybrid multicenter clinical trial. Speech recognition scores taken six months after CI activation were 64% (CNC in quiet) and 48% (AzBio in +5 dB SNR) in the EAS mode (Roland et al., 2016).

All participants used the N6 processor, which was programmed using Custom Sound (clinical programming software provided by Cochlear Ltd). The acoustic component was fit using the National Acoustic Laboratories-Non Linear 2 fitting formula (NAL-NL2; Keidser et al., 2011) and verified via real ear measures. The electrical component was programmed with the Advanced Combination Encoder processing strategy (ACE; Skinner et al., 2002). Custom Sound incorporated the patient's unaided audiogram to suggest acoustic and electric cutoff frequencies, which were sometimes altered based on the programming clinician's judgment. The average low-pass cutoff frequency for acoustic stimulation and high-pass cutoff frequency for electrical stimulation are shown in Fig. 1(right) and are based on an average of cutoffs listed in Table I. The acoustic cutoff frequency was considered as the highest audiometric frequency that matched NAL-NL2 targets within ±6 dB, based on real ear measures. The electric cutoff frequency was derived directly from the subject's MAP. (Note that subject L82R's acoustic component was 10–20 dB under target for 250–500 Hz; however, she was not reprogrammed given that she was not a patient of our clinic and she was accustomed to her current programming.) Subjects, on average, received information acoustically up to 667 Hz and electrically from 583 Hz and above, indicating some frequency overlap between acoustic and electric stimulation. The average acoustic cutoff of 667 Hz reflects UIHC's programming philosophy of not providing amplification for hearing losses >70 dB hearing level (HL; see the average audiogram plot in Fig. 1) given the limited benefits of amplification for such severe losses (Turner, 2006). For electrical hearing, Custom Sound by default deactivates the first four basal electrodes (electrodes 1–4) for the L24 array because these basal electrodes are usually too lateral relative to the modiolus/auditory nerve to provide effective electrical stimulation. Other electrodes were deactivated as needed to address sound quality issues, non-auditory sensations, and open/short circuits (Table I).

C. Speech masking release

Speech recognition was evaluated via a 12-alternative-forced-choice (AFC) closed-set spondee identification task with feedback provided. The spondees were digitized recordings of a male talker (Harris, 1991) with an average f₀ of 107.83 ± 3.36 Hz, calculated using Praat (Boersma and Weenink, 2017). Spondee identification was assessed in quiet, two-female talker babble (f₀ = 190.05 ± 0.54 Hz), and ten-female talker babble (f₀ = 190.13 ± 3.25 Hz). A female talker was used to provide voice pitch cues and promote segregation of the target from the masker, which can contribute to masking release (Stickney et al., 2004). Figure 2(left) shows the frequency spectra of the spondees and babble masker, which illustrate the differences in low-frequency energy, the energetic masking of the spondees in the mid-to-high frequencies, and the similarities in spectra for both babble maskers. Figure 2(right) shows the time waveforms of the babble maskers. The two-talker babble contains prominent amplitude variations and was used as the “modulated” masker. The ten-talker babble was used as the relatively “steady” masker because the addition of more talkers leads to a flatter temporal envelope (albeit still with some temporal fluctuations) and greater masking (Miller, 1947; Simpson and Cooke, 2005). To generate the babble stimuli, 12 freely available audiobooks from 12 female speakers were downloaded from LibriVox.² Adobe Audition CC 2017 (San Jose, CA) was used to extract 10-s clips from each audiobook, taking care to ensure that the extracted clips had the same f₀ across clips (verified by Praat). Waveforms were subsequently root-mean-square (RMS)-normalized and combined to create a two-talker babble and a ten-talker babble.

FIG. 2.

(Color online) Frequency spectra (left) and time waveforms (right) for spondee and babble stimuli. The spondee spectrum is an average of the frequency magnitude spectra of all 12 spondees. The amplitude of the spondee in the time domain was reduced by 5 dB prior to frequency analysis, reflecting the −5 dB SNR test condition used for this study. The vertical lines indicate the average high-pass cutoff frequency for electrical stimulation and the average low-pass cutoff frequency for acoustic stimulation. Note that in the time waveforms the two-talker babble (top) has a peakier waveform relative to the ten-talker babble (bottom).

Subjects completed a practice run consisting of 2 presentations of the 12 spondees (for a total of 24 presentations) presented in quiet, two-talker, and ten-talker babble. The starting location of the babble was randomized across trials to minimize local spectral and temporal dips in the masker as potential cues. The onset of the babble was 1.5 s prior to the spondee while the offset coincided with the spondee. The test run consisted of 4 presentations of the 12 spondees (48 total presentations) in two-talker babble and another set of 4 presentations in ten-talker babble. Spondees were presented at −5 dB SNR, based on data showing SRTs in noise/babble ranging from −11 dB (better) to −4 dB (worse) for S8 Hybrid CI users (Turner et al., 2004; Golub et al., 2012) and knowledge that masking release is seen in difficult (i.e., negative) SNRs (Bernstein, 2011, for review).

D. Spectral ripple density discrimination

Spectral ripple density discrimination thresholds were measured via a three-AFC, two-up, one-down adaptive paradigm (Levitt, 1971). The ripple density, measured in ripples per octave (RPO), was varied to determine the highest ripple density that could be discriminated.

The stimuli were a sum of 1976 pure tones from 100 to 8000 Hz in 4 Hz steps. The starting phase of each pure tone was randomized to minimize fine structure cues. The amplitude of each pure tone was calculated as in Litvak et al. (2007):

$$A\left(f\right)={10}^{(d/2)\{\hspace{0.17em}\text{sin}\hspace{0.17em}\left[2\pi \hspace{0.17em}\hspace{0.17em}{\text{log}}_2\left(f/f_{\text{low}}\right)\text{RPO}+\theta \right]/20\}}.$$ (1)

A(f) represents the amplitude of a pure tone with frequency f, d represents the ripple depth in dB, f_low represents the lower frequency bound, RPO represents the ripple density, and θ represents the starting ripple phase. The ripple depth (d) was fixed at 30 dB, and the lower frequency bound (f_low) was set to 100 Hz. The starting ripple phase was randomized (as detailed below). Stimulus duration was 500 ms, including a 50 ms onset/offset tapered by a Hanning window. Stimuli were also bandpass filtered between 100 and 8000 Hz to reduce spectral edges (supplemental Fig. 1).¹

In the adaptive procedure, both the reference and test intervals contained a ripple stimulus with identical ripple density but differing starting phases. For each trial, the ripple starting phase (θ) was randomized between 0 and 2π in the reference intervals with both reference stimuli having the same starting phase. The starting ripple phase of the test stimulus was offset by π/2 relative to that of the standard stimulus, resulting in an inverted stimulus (supplemental Fig. 1).¹ The trial-by-trial randomization of ripple starting phase minimized local cues related to spectral peak locations. In addition, the stimuli were generated online, so the test and reference intervals contained a fresh instance of the stimulus, further minimizing any fine structure cues that could arise had the ripple starting phase and the pure tone starting phase been fixed. Step sizes of ½RPO were utilized. Testing terminated after ten reversals with the threshold based on the average of the last six reversals. The average of three runs was used to determine the final threshold. The stimuli were presented at 65 dBA with a ±3 dB level rove and a 1-s interstimulus interval.

E. f₀DLs

f₀DLs were measured via a three-AFC, two-down, one-up adaptive paradigm (Levitt, 1971). The Δf₀ between two harmonic complexes was varied to determine the minimum Δf₀ that could be discriminated between two complexes.

The stimulus was a harmonic complex with f₀ = 110 Hz and a frequency range spanning from 110 to 7920 Hz (supplemental Fig. 2).¹ This f₀ is identical to that of the male spondees used for the masking release task. Stimulus duration was 500 ms, including a 50 ms onset/offset tapered by a Hanning window.

In the adaptive procedure, the test interval contained a complex with f₀ = (110 + Δf₀) Hz, where the starting f₀ was initially set 20% higher than the reference stimulus, and was changed by a factor of 0.50 for the first four reversals and by a factor of 0.25 for the final six reversals. In line with previous studies (Shackleton and Carlyon, 1994; Moore et al., 2006a,b; Oxenham et al., 2009), the reference f₀ was roved by 10%. As a result, the reference f₀ of each trial varied from 99 to 121 Hz. The trial-by-trial randomization of f₀ minimized the listeners' reliance on an internal memory of the reference stimulus had the f₀ been fixed. The average of three runs was used to determine the f₀DLs. Stimuli were presented at 65 dBA with a ±3 dB level rove and a 1-s interstimulus interval.

F. Statistical analysis

Raw experimental data are supplied as Table 2 of the supplementary material. Statistical analysis was performed using SPSS 22 (IBM, Armonk, NY). When linear mixed effects (LME) modeling was used, the dependent variable was the fixed effect with subjects entered as the random effects. The model did not assume equal variances across test conditions. The significance level was established at α = 0.05, and Bonferroni corrections were applied to adjust p-values for follow-up multiple pairwise comparisons as warranted. Given that hypotheses 2–4 are one-directional and correlational, simple linear regression in combination with one-tailed hypothesis testing was used to calculate the statistical significance of regression fits.

III. RESULTS

A. Clinical outcomes

Speech recognition performance in quiet (CNC words; 60 dBA) and multi-talker babble (AzBio sentences; 60 dBA; +5 dB SNR) was obtained from clinical records (Fig. 3). The LME analysis indicated a significant effect of listening mode on CNC scores (F_2,19.485 = 81.208, p < 0.001). Follow-up pairwise comparisons showed CNC scores significantly improved from 20 ± 14% preoperatively to 53 ± 20% in the E-Only mode (p < 0.001) and to 72 ± 16% in the EAS mode (p < 0.001). Significant improvement was also noted with EAS compared to the E-Only listening mode (p < 0.001). LME analysis also indicated a significant effect of listening mode on AzBio scores (F_2,10.603 = 17.400, p < 0.001). Pairwise comparisons indicated AzBio scores significantly improved to 43 ± 24% with EAS compared to 14 ± 15% preoperatively (p < 0.001) and E-Only (20 ± 16%, p < 0.001). There was no significant difference between preoperative and E-Only scores (p = 0.496), indicative of the role of residual acoustic hearing in speech understanding in noise.

FIG. 3.

(Color online) Clinical preoperative and postoperative speech recognition outcomes. (Left) CNC word scores in quiet and (right) AzBio sentence scores in noise. Preoperative indicates aided scores in the ear to be implanted. E-Only indicates electric-only. EAS indicates combined electric and acoustic stimulation. Open squares indicate mean ±1 SD.

B. Speech masking release

Figure 4 shows spondee recognition performance in quiet for each listening mode. Performance was relatively high in the A-Only condition (77 ± 22%) and approached ceiling for the E-Only and EAS conditions (98 ± 4% and 100 ± 1%, respectively). LME analysis indicated a significant effect of listening mode (F_2,19.166 = 11.393, p = 0.001). Pairwise comparisons indicated significant improvement with the use of E-Only hearing relative to A-Only hearing (p = 0.001) and with use of EAS relative to A-Only hearing (p < 0.001). There was no significant difference between E-Only and EAS (p = 0.213).

FIG. 4.

(Color online) Spondee recognition in quiet for E-Only, A-Only, and EAS modes. Open squares indicate mean ± 1SD. The dashed line represents chance performance.

Spondee recognition in ten-talker and two-talker babble for each listening mode is plotted in Fig. 5 to show presence/absence of masking release. A LME analysis with listening mode (E-Only, A-Only, EAS) and babble type (two-talker, ten-talker) as fixed effects and subjects as random effects was performed to analyze the effects of listening mode and babble type on spondee recognition. The analysis indicated a significant interaction between listening mode and talker type (F_2,19.133 = 23.891, p < 0.001). Follow-up pairwise comparisons of spondee recognition in two-talker vs ten-talker babble for each listening mode show no significant masking release for E-Only hearing (4 ± 8%, p = 0.570), whereas A-Only and EAS hearing lead to significant masking release (13 ± 10%, p < 0.001 and 17 ± 8%, p < 0.001, respectively), indicative of the importance of acoustic hearing in this task. Note that the A-Only and EAS masking releases observed are also above chance performance (8.33%). Additionally, follow-up pairwise comparisons of spondee recognition in two-talker babble for each listening mode showed the performance in EAS (60 ± 23%) was significantly better than that in A-Only (44 ± 21%, p = 0.028) and E-Only modes (23 ± 10%, p < 0.001). The performance in A-Only was also significantly better than that in E-Only (p = 0.006). Follow-up pairwise comparisons of spondee recognition in ten-talker babble for each listening mode showed the performance in EAS (43 ± 19%) was significantly better than that in A-Only (31 ± 18%, p = 0.022) and E-Only (19 ± 8%, p < 0.001). The performance between A-Only and E-Only modes was not significantly different (p = 0.248).

FIG. 5.

(Color online) Spondee recognition in two-talker vs ten-talker babble for E-Only, A-Only, and EAS modes. Open squares indicate mean ±1 SD. The dashed line represents chance performance.

Each subject's masking release for each listening mode was calculated from the data presented in Fig. 5 and plotted in Fig. 6(left). LME analysis revealed a significant effect of listening mode (F_2,19.161 = 23.884, p < 0.001). Follow-up pairwise comparisons showed that compared to masking release in the E-Only mode (4 ± 8%), masking release significantly improved in the A-Only and EAS listening modes (13 ± 10%, p = 0.006 and 17 ± 8%, p < 0.001, respectively). Moreover, masking release was not significantly different between the A-Only and EAS modes (p = 0.337), indicative of the role of residual acoustic hearing in masking release. One confound in this analysis is that baseline spondee recognition scores in quiet differ across subjects and listening modes (Fig. 4). The difference in speech recognition scores in quiet and in a steady masker represents the best possible masking release that can be obtained. Thus, a normalized masking release was calculated to quantify how close to the maximal masking release each subject obtained [Fig. 6(right); Jin and Nelson, 2006]:

$$\text{Normalized}\hspace{0.17em}\text{masking}\hspace{0.17em}\text{release}\hspace{0.17em}\left(\%\right)=\frac{{\text{Modulated}}_{\text{Score}}-{\text{Steady}}_{\text{Score}}}{{\text{Quiet}}_{\text{Score}}-{\text{Steady}}_{\text{Score}}}\ast 100.$$ (2)

There was a significant effect of listening mode on normalized masking release (F_2,18.171 = 53.323, p < 0.001). Follow-up pairwise comparisons indicated that compared to the E-Only mode (5 ± 10%), normalized masking release improved significantly in the A-Only (31 ± 18%, p < 0.001) and EAS modes (36 ± 23%, p < 0.001) and was not significantly different between the A-Only and EAS modes (p = 1.000), indicative of the role of acoustic hearing. (Subject L74L's A-Only normalized masking release of −57% was a clear outlier and was removed from the analysis. However, LME and follow-up pairwise analyses were also redone with this score included, and they produced similar results.) The trends in the normalized data are similar to the raw masking release data in that the E-Only performance was poor, whereas the A-Only and EAS performances were similarly superior. The advantage of the normalized data is that it highlights the extent to which subjects took advantage of gaps in modulated noise regardless of baseline recognition scores in quiet.

FIG. 6.

(Color online) The masking release and normalized masking release for the E-Only, A-Only, and EAS modes. Open squares indicate mean ± 1SD. The dashed line represents no masking release.

C. Psychophysics

Spectral ripple density discrimination thresholds and f₀DLs were assessed in E-Only, A-Only, and EAS listening modes (Fig. 7). Ripple density discrimination thresholds were poorest for the E-Only listening mode (1.78 ± 0.85 RPO) and improved in A-Only and EAS listening modes (2.89 ± 1.44 and 2.83 ± 1.03 RPO, respectively). LME analysis indicated a significant effect of listening mode (F_2,18.02 = 15.21, p < 0.001). Follow-up pairwise comparisons indicated that the improvement was significant in the A-Only and EAS conditions relative to the E-Only condition (p = 0.003 and p < 0.001, respectively). There was no significant difference between the A-Only and EAS conditions (p = 1.000).

FIG. 7.

(Color online) Spectral ripple density discrimination thresholds and f₀DL performance for the E-Only, A-Only, and EAS listening modes. The open squares indicate mean ±1 SD.

There was no significant effect of listening mode on f₀DLs (F_2,18.23 = 2.045, p = 0.158), likely a consequence of the high variance in E-Only f₀DLs. However, average performance trends were similar to the spectral ripple performance; f₀DL performance in the E-Only mode was poor (20.96 ± 39.56 Hz) but improved for the A-Only and EAS listening modes (5.43 ± 4.56 Hz and 5.37 ± 3.36 Hz, respectively). Similarities in the A-Only and EAS modes imply that acoustic hearing is a dominant factor in EAS listening for both spectral ripple density discrimination and f₀DL performance.

D. Speech masking release vs psychophysics

We hypothesized a relationship between spectral resolution (measured via spectral ripple density discrimination), f₀/pitch sensitivity (measured via f₀DL testing), and speech masking release. The correlations presented focus on performance with EAS because it is the everyday listening mode. Lower f₀DLs were inversely and significantly correlated with speech masking release [r₂₀ = −0.453, p = 0.017, Fig. 8(left)], and spondee recognition in two-talker and ten-talker babble (r₂₀ = −0.557, p = 0.004 and r₂₀ = −0.489, p = 0.010, respectively; data are not shown). No correlation was observed between spectral ripple density discrimination thresholds and speech masking release [r₂₀= −0.004, p = 0.492; Fig. 8(right)] and spondee recognition in two-talker and ten-talker babble (r₂₀ = 0.231, p = 0.150 and r₂₀ = 0.288, p = 0.097, respectively; data are not shown).

FIG. 8.

(Color online) Correlations between speech masking release and psychophysical performance in the EAS listening mode.

E. Influence of residual hearing on speech masking release and psychophysics

The third hypothesis was that greater residual hearing (PTA of 125, 250, and 500 Hz) would correlate with higher spectral ripple density discrimination thresholds, lower f₀ DLs, and greater masking release in the EAS mode. However, no correlations between residual hearing and any of these measures were observed (spectral ripple density discrimination, r₂₀ = −0.277, p = 0.106; f₀DL, r₂₀ = 0.145, p = 0.259; speech masking release, r₂₀ = 0.019, p = 0.467; data are not shown).

F. Relationship between speech masking release and clinical outcomes

Postoperative AzBio recognition scores in noise were significantly correlated with speech masking release in the EAS mode (Fig. 9; r₂₀ = 0.473, p = 0.013), which appears to indicate that the ability to listen in the dips relates to how well one does on open-set speech recognition in noise.

FIG. 9.

(Color online) Correlation between AzBio sentence recognition in noise (+5 dB SNR) and speech masking release in the EAS listening mode.

IV. DISCUSSION

Speech recognition in quiet was significantly better with the E-Only and EAS modes compared to the pre- and postoperative measures conducted using only acoustic hearing (Figs. 3 and 4). In contrast, E-Only hearing is less helpful for speech recognition in noise; acoustic hearing was necessary to provide significant postoperative benefit in noise (Figs. 3 and 5). Similarly, despite the use of negative SNRs and closed-set testing to promote masking release (Buss et al., 2009; Bernstein, 2011), there was minimal release with electric hearing (Fig. 5 and 6). This is consistent with prior masking release studies in standard CI users, despite differences in speech and masker stimuli and the use of open- vs closed-set testing paradigms (Nelson et al., 2003; Stickney et al., 2004; Turner et al., 2004; Fu and Nogaki, 2005; Won et al., 2007; Jin et al., 2013). Some reports suggest modulation detection interference in electric hearing causes difficulty parsing the competing modulations of the speech stimuli and the modulated masker, which counteracts dip-listening and leads to minimal masking release or, in some cases, worse performance in modulated noise vs steady noise (Kwon and Turner, 2001; Kwon et al., 2012; Jin et al., 2013). The similar temporal modulations in the two-talker babble and the target spondee may have caused interference effects in the present study, leading to a slightly worse performance in two-talker babble in electric hearing for some subjects (Figs. 5 and 6). In contrast, previous studies show noise modulated at higher rates leads to greater masking release (Nelson et al., 2003; Fu and Nogaki, 2005; Jin et al., 2013). Electrical current spread in the cochlea can also cause overlapping temporal fluctuations to sum. This smooths the temporal envelope and minimizes differences between steady and modulated noise, leading to minimal masking release (Oxenham and Kreft, 2014; also see Stone et al., 2011; Stone et al., 2012).

EAS was necessary to achieve a significant clinical benefit in noise (Fig. 3). Likewise, although acoustic hearing was not sufficient for spondee recognition in quiet (Fig. 4), it contributed significantly to spondee recognition in noise and masking release for the A-Only and EAS modes (Figs. 5 and 6). Furthermore, masking release was not significantly different between the A-Only and EAS modes and was significantly greater than the E-Only mode (Fig. 6), suggesting that release was primarily driven by acoustic hearing. These findings highlight the limitations of electric hearing and advantages of acoustic hearing in noise. They are consistent with prior reports of EAS CI users showing a greater acoustic benefit in noise relative to quiet when tested in EAS vs E-Only hearing on clinical speech recognition tasks (Gstoettner et al., 2008; Helbig et al., 2011; Kim et al., 2017). Results are also consistent with previous reports of improved speech masking release in EAS simulations compared to fully vocoded speech and in EAS CI users (Turner et al., 2004; Hopkins and Moore, 2009). Furthermore, current results help separate the contributions of A-Only, E-Only, and EAS hearing to masking release.

The lack of low-frequency electric stimulation in the E-Only mode may confound the interpretation that acoustic hearing is necessary for speech recognition in noise and for masking release. However, even EAS patients with full bandwidth electric stimulation show speech recognition improvements when low-frequency acoustic amplification is provided in the implanted ear (Fraysse et al., 2006; Gstoettner et al., 2008; Simpson et al., 2009; Helbig et al., 2011; Gifford et al., 2017). In addition, standard CI users with a full frequency allocation do not show masking release (Nelson et al., 2003; Stickney et al., 2004; Turner et al., 2004; Fu and Nogaki, 2005; Won et al., 2007; Jin et al., 2013). A future experimental manipulation could involve programming the electrical component to provide a full bandwidth signal for comparisons between E-Only and EAS modes. However, compressing an entire frequency range into an already short electrode further disrupts tonotopic matching (Landsberger et al., 2015) and may diminish speech understanding (Başkent and Shannon, 2004, 2005). This may explain why some EAS CI users demonstrate better speech recognition with restricted high-frequency electric stimulation rather than full bandwidth electric stimulation (Fraysse et al., 2006; Vermeire et al., 2008; Karsten et al., 2013; Gifford et al., 2017). Theoretically, such distortions introduced by compression may reduce masking release in EAS as well.

The significant correlation between f₀DLs and speech masking release in the EAS mode in combination with poor E-Only f₀DLs and better A-Only and EAS f₀DLs suggests a role of f₀ encoding by low-frequency acoustic hearing (Figs. 7 and 8). f₀ information conveyed by low-frequency TFS and spectral cues likely aided segregation of the target from the masker, as well as glimpsing the speech within the fluctuations of the masker, leading to masking release. This is consistent with prior speech-on-speech masking paradigms showing poorer f₀DLs correlating with poorer speech recognition in NH and HI listeners (Glasberg and Moore, 1989; Summers and Leek, 1998). The restricted frequency allocation of the E-Only mode also prevents encoding of f₀, which may contribute to poor f₀DLs in E-Only stimulation and improved f₀DLs in the A-Only and EAS modes (Fig. 7). However, NH listeners tested using vocoded speech and CI users do not show improvements in speech understanding when f₀ differences between competing speech are introduced, even though the full frequency allocation is provided (Qin and Oxenham, 2005; Stickney et al., 2007; Arehart et al., 2011; Auinger et al., 2017). Additionally, E-Only f₀DLs were consistent with previous reports of standard CI users (Geurts and Wouters, 2001; Zeng, 2002; Goldsworthy et al., 2013; Goldsworthy, 2015) and NH listeners using vocoded stimuli (Qin and Oxenham, 2005; Schvartz-Leyzac and Chatterjee, 2015). The f₀DLs obtained in the A-Only and EAS modes were similar to previous reports in HI listeners (Glasberg and Moore, 1989; Moore et al., 2006b; Vongpaisal and Pichora-Fuller, 2007). Due to the better discrimination of f₀ cues via acoustic hearing, benefits of increasing f₀ differences between the target and the masker are shown for listeners with hearing loss (Arehart et al., 1997), EAS CI simulations (Arehart et al., 2011), and EAS CI users (Auinger et al., 2017).

Despite the improved ripple density discrimination thresholds with acoustic hearing suggesting improved spectral resolution, discrimination thresholds were not correlated with masking release in the EAS mode (Figs. 7 and 8). This was an unexpected outcome given that increasing the number of spectral bands in vocoded speech has resulted in greater masking release (Nelson et al., 2003; Qin and Oxenham, 2003; Nelson and Jin, 2004; Fu and Nogaki, 2005). However, the lack of significant correlation between ripple thresholds and masking release may reflect recent concerns that the ripple task may not adequately reflect spectral resolution (despite Anderson et al., 2011, correlating ripple results to psychophysical tuning curves). A potential confound in A-Only ripple thresholds is that the ripple stimulus is low-pass filtered due to the sloping hearing loss. The stimulus energy passing through the filter differs depending on the ripple phase, leading to potential local intensity and spectral edge cues between standard and inverted stimuli (also see Azadpour and McKay, 2012, who raised similar concerns in electric hearing). With respect to electric hearing, the wide filter bandwidths assigned for each electrode likely cannot encode densely spaced ripples, leading to distortions and aliasing in the processor output, particularly at high ripple densities (>2 RPO; O'Brien and Winn, 2017). Using an average EAS audiogram (Fig. 1), the average electrical frequency range is 563–7938 Hz over electrodes 5–22 in L24 Hybrid CI users (recall E1–E4 are deactivated by default in the programming software). Whereas a standard Cochlear CI frequency MAP will assign 563–7938 Hz to electrodes 4–22 (with 188–563 Hz allocated to electrodes 1–3). Thus, the filterbands are slightly wider in the electric portion for L24 Hybrid CI users, which may slightly decrease the upper limit of the ripple encoding. Widened acoustic filters in HI listeners (Peters and Moore, 1992) also pose a theoretical limitation in acoustic encoding of spectrally dense ripple stimuli in A-Only and EAS hearing as well.

With respect to potential local intensity and edge cues, CI users' spectral ripple discrimination scores are not affected by relatively large level roves (±15 dB), the use of random vs fixed starting ripple phase, nor the use of spectrally windowed stimuli (Anderson et al., 2011; Won et al., 2011). These manipulations, designed to reduce local intensity cues and edge effects, were used in our setup. However, a spectral modulation depth detection task or a spectral-temporal ripple task would further reduce extraneous cues and still correlate with speech recognition in HI and CI groups (Aronoff and Landsberger, 2013; Gifford et al., 2014; Gifford et al., 2018; Mehraei et al., 2014; Bernstein et al., 2016) and should be considered in future experimental paradigms.

Nevertheless, E-Only ripple density discrimination thresholds were consistent with average thresholds of 0.62–1.73 RPO reported in standard CI users, despite differences in spectral ripple stimuli and the lack of low-frequency stimulation in E-Only listening (Fig. 7; Henry et al., 2005; Won et al., 2007; Anderson et al., 2011; Jeon et al., 2015). Thresholds in the A-Only and EAS modes span the range of previously reported average thresholds for HI listeners (1.77–3.25 RPO; Henry et al., 2005; Davies-Venn et al., 2015) and EAS CI users (4.60 RPO; Golub et al., 2012). Anderson et al. (2011) demonstrated individual across-electrode variations in CI listeners' ripple discrimination thresholds using band limited ripple stimuli targeting localized regions of the electrode array. They correlated the best localized ripple discrimination threshold with the broadband ripple discrimination threshold, possibly indicating that broadband discrimination thresholds reflect cochlear regions where peripheral health is best. Their results, in addition to Golub et al. (2012) and trends in ripple and f₀DL tasks (Fig. 7), demonstrate that EAS CI users rely on localized regions of good peripheral health whether provided by acoustic or electric hearing.

Contrary to our hypothesis, no correlations were seen between low-frequency PTAs (125–500 Hz) and speech masking release, spectral ripple density discrimination, and f₀DLs. Although prior reports have shown a correlation between residual acoustic hearing and EAS benefit in noise (Turner et al., 2008; Gifford et al., 2013), the null finding in our case was not necessarily surprising. Audiometric thresholds do not always reflect deficits in speech recognition in noise (e.g., Grant and Walden, 2003) or reduced spectral/temporal resolution (e.g., cochlear “dead regions”; Moore, 2004).

A closed-set spondee recognition task is relatively simple with easily recognized words, and one might argue it may not reflect clinical assessments of speech understanding in noise. However, the significant relationship between masking release and AzBio sentence recognition in babble, as assessed in the EAS mode (Fig. 9), implies that the ability to listen in the dips relates to real-world open-set speech recognition in noise. Given the strong correlation, there is potential clinical utility in using masking release to evaluate benefits of hearing preservation beyond preserving audiometric thresholds. The nature of closed-set testing allowed evaluation of acoustic and electric contributions to masking release. It is possible that patients who do not show increased release with acoustic hearing may benefit from changes in implant programming to optimize the use of acoustic and/or electric hearing for speech understanding, such as providing a greater bandwidth of acoustic amplification.

A. Caveats of the masking release paradigm/future directions

1. Psychometric functions

Psychometric functions for speech recognition performance in steady noise are steeper than performance in modulated noise; thus, the masking release obtained at a given SNR is partially due to slope differences (Stickney et al., 2004; Bernstein and Grant, 2009; Hopkins and Moore, 2009; Bernstein and Brungart, 2011; Shen et al., 2015; but see Qin and Oxenham, 2003). Slopes could potentially differ across the A-Only, E-Only, and EAS modes (Hopkins and Moore, 2009), which could confound interpretations of acoustic and electric contributions to masking release. Future studies could obtain psychometric functions for the A-Only, E-Only, and EAS modes so that masking release could be interpreted by considering the differences in psychometric functions.

2. Stimuli considerations

The type of masker used can affect masking release (Stickney et al., 2004; Croghan and Smith, 2018). A multi-talker masker was chosen because it contains dynamically varying spectral and temporal peaks and is more reflective of everyday listening environments. Spectral resolution and TFS may be more critical in this listening situation (e.g., Croghan and Smith, 2018). However, informational masking can occur with talker backgrounds, and the relative contributions of energetic and informational masking could vary, depending on the number of talkers used (Brungart et al., 2001; Simpson and Cooke, 2005). Steady-state and modulated broadband noise maskers avoid informational masking, but inherent temporal fluctuations in noise can create additional masking effects beyond energetic masking (Stone et al., 2011; Stone et al., 2012) and cause further difficulties in detecting the dips of a fluctuating masker. Alternatively, low-fluctuation noise (Pumplin, 1985) could be used to evaluate masking release in EAS CI users as done in NH listeners (Stone et al., 2011; Stone et al., 2012). A more complete picture of masking release in EAS populations could be obtained using different types of maskers—babble, noise, and low-fluctuation noise—to ascertain contributions of informational, energetic, and modulation masking to masking release.

The lack of significant correlation between ripple density discrimination thresholds and masking release may indicate that spectral resolution is not as critical for a closed-set task (assuming ripple thresholds are truly reflective of spectral resolution). For example, Davies-Venn et al. (2015) reported trends approaching significance between spectral ripple density discrimination thresholds and open-set sentence recognition in noise for HI listeners. For the current study, informal conversations with some participants revealed that they did not always understand the entire spondee, but they glimpsed a small portion of the spondee within the valleys of the masker and matched it with 1 of the 12 possible responses. Spectral resolution may play a bigger role had more difficult but still closed-set stimuli been used (e.g., “double-spondees”; see Davies-Venn et al., 2011; or Coordinate Response Measure stimuli; see Bolia et al., 2000).

3. Outliers

Participants L74L, L59R, and L6R may be outliers in the f₀DL correlations (Fig. 8). The possible outliers, as well as the lack of correlation between speech masking release and the spectral ripple discrimination task, may be due to subject sampling and high variance. Correlations and variance in the current study can change with a larger subject pool. For example, even with a large CI population (100+ ears) leading to highly significant correlations (p < 0.0001), spectral ripple depth detection accounts for 25%–66% of the variance in speech recognition in quiet and noise (Gifford et al., 2014; Gifford et al., 2018).

The wide distribution of ages in our subject group (21–78 years old) could lead to potential age effects in speech perception and psychophysical performance (Dubno et al., 2002; Schvartz et al., 2008; Schvartz-Leyzac and Chatterjee, 2015; Sladen and Zappler, 2015; but see Takahashi and Bacon, 1992; Souza and Turner, 1994; Mussoi and Brown, 2019). Participants L74L, L59R, and L6R are also older, which could explain trends seen in f₀DL correlations. Focusing on EAS listening, older subjects did show poorer f₀DLs (r₂₀ = 0.480, p = 0.012), masking release (r₂₀ = −0.3736, p = 0.043), and normalized masking release (r₂₀ = −0.4635, p = 0.015). There were also trends of poorer CNC scores (r₂₀ = −0.3283, p = 0.068) and AzBio scores (r₂₀ = −0.301, p = 0.087), and a less convincing trend of poorer spectral ripple density discrimination thresholds (r₂₀ = −0.220, p = 0.163).

V. CONCLUSIONS

Speech masking release was demonstrated in EAS CI users with the use of acoustic hearing. Significant correlations with f₀DLs and the lack of correlation with spectral ripple density discrimination indicate a likely role of TFS cues in acoustic hearing. The access to TFS appeared to help listeners exploit f₀ differences between target and masker as well as extract speech within the fluctuations of the masker. The role of spectral resolution cannot be completely excluded given caveats in the spectral ripple density paradigm and the role of spectral resolution in f₀ encoding. The significant correlation with AzBio sentence recognition in noise indicates potential clinical utility of the speech masking release task to evaluate/optimize the use of EAS hearing.