Noise Management Preferences During Long-Term Hearing Aid Usage and Their Relation to Audiologic Factors

Participants

The study included 123 experienced HA users (52 female, 71 male; mean age 65.2 years), native speakers of German (n = 82) or Japanese (n = 41). The average age was 65.2 years (standard deviation (SD): 11; range: 29–80) with only minor deviations in the two sub-populations (German: 65.2 years; Japanese: 65.0 years). All participants had at least 3 months of HA experience (average 7 years) and met national guidelines for HA provision. Exclusion criteria were cochlear-implant candidacy, the primary reason for HA provision being tinnitus management, suspected cognitive impairment, difficulty operating the devices, history of severe head injury, or enlarged ear canals from prior otologic surgery. Data collection in Germany ran from April 2022 to October 2023 (18 months), and in Japan from February 2022 to December 2024 (35 months).

Pure-Tone Audiometry

All participants underwent otoscopy and standard clinical audiometry (Germany: Affinity 2.0, Interacoustics; Japan: AA-H1 or AA-M1A, Rion), with air-conduction thresholds measured from 0.25 to 8 kHz. Figure 1 shows the pure-tone thresholds for the right and left ears, representing hearing losses ranging from mild to severe. One hundred and nine participants showed across-ear threshold differences ≤15 dB for at least four out of the six measured audiometric frequencies, indicating rather symmetric hearing losses, whereas the remaining 14 participants exceeded that limit, indicating a higher degree of asymmetry. Hearing thresholds were summarized using the four-frequency binaural pure-tone average (BPTA4: average pure-tone thresholds across ears and frequencies of 0.5, 1, 2, and 4 kHz), which was used as a predictor in the statistical analysis of participant preferences.

OPEN IN VIEWER

Figure 1.

Pure-tone thresholds obtained for the right and left ears of the 123 individual participants (thin lines) along with the mean (bold line) and standard deviation (shaded area) across participants.

Hearing-Aid-Fitting Procedure

Participants were fitted with Oticon More 1 minireceiver-in-the-ear HAs, a premium model released in 2020. Participants were fitted according to the standard clinical practices in their respective countries:

The 82 German participants were fitted according to the NAL-NL2 rationale for experienced users (Keidser et al., 2011) by certified hearing-care professionals using Oticon Genie 2. Their acoustic coupling followed Genie 2 recommendations, resulting in a wide range of fittings, from open bass domes to custom earmolds with varying vent sizes (see Jürgens et al., 2025 for details). Amplification was verified using real-ear measurement (REM) in Affinity 2.0 and adjusted via REM AutoFit in Genie 2 with the international speech test signal (Holube et al., 2010).

The 41 Japanese participants were fitted according to the Utsunomiya method (Shinden et al., 2021; Yamada et al., 2020) and the Japan Audiological Society guidelines (Kodera et al., 2016). Following local practice, they received non-vented custom earmolds, and amplification was assessed using Rion AA-H1 or AA-M1A analyzers. REM verification showed that the Utsunomiya method typically prescribed 5–10 dB more gain than NAL-NL2 from 0.5 to 4 kHz, to accommodate Japanese acoustic properties (Suzuki et al., 2023), which rely heavily on vowel audibility due to the predominance of consonant–vowel syllables.

The different DIR + NR settings were configured using Oticon's standard fitting software (Genie 2), as described in section HA DIR + NR feature. Seven noise programs (NPs) were available, with the strength of DIR + NR increasing as the NP increased.

Audiologic Measures

Audiologic measures collected during the initial visits included aided speech reception thresholds (SRTs), real-ear occluded insertion gain (REOIG), as described in Jürgens et al. (2025) and audible contrast threshold (ACT™) values as described in ZaarSimonsen et al. (2024).

SRTs were measured using the everyday sentences material of the hearing in noise test (HINT) in the participants’ native language (German: Joiko et al., 2020; Japanese: Shiroma et al., 2008). Sentences were presented from the front at 65 dB SPL, while competing speech maskers spoken by two different male talkers, mixed with low-level speech-shaped noise and presented 6 dB higher than the noise, were played from two lateral loudspeakers (at ±100°). The level of the maskers was adaptively adjusted to determine the SNR at which 50% of the sentences were correctly repeated. Testing was performed with different HA processing settings, as well as without HA in the initial session. To allow comparison between German and Japanese participants, the SRTs measured in the HINT were corrected for normative differences between the German and Japanese HINT by subtracting 1.3 dB from the SRTs collected for the Japanese population, as also done in Jürgens et al. (2025).

REOIG was measured for both ears of each participant as proposed by Cubick et al. (2022). For each participant, sound levels in the ear canal were measured without a HA and then again with the HA inserted but switched off. The difference between these two measurements across frequency reflected the degree to which the ear canal was acoustically occluded by the coupling. To obtain a single measure of acoustic coupling closedness, the REOIG was interpolated to the frequencies specified in ANSI S3.5-1997, Table 1. Each interpolated REOIG value was then frequency-weighted using the band-importance values of the Speech Intelligibility Index (SII; 1997, Table 1) and summed, yielding a speech-weighted average REOIG for each ear. These ear-specific averages were subsequently averaged across ears to obtain an individual measure of acoustic coupling closedness. REOIG values are typically negative, with more negative values indicating a more closed acoustic coupling.

OPEN IN VIEWER

Table 1.

Definition of NPs in Noise and Corresponding DIR and NR Settings.

NP	DIR	NR	Description
1	Omnidirectional with pinna compensation only	Off	No noise management in noise
2–6	Minimum-variance distortionless-response directionality	DNN based	Automatically steered DIR and NR activation strength increases with NP
7	Front-facing directionality	DNN based	Maximum noise management

DIR = directionality; NR = noise reduction; DNN = deep neural network; NP = noise program.

ACT was used to assess listeners’ sensitivity to spectro-temporal modulation, with higher values indicating poorer sensitivity (ZaarSimonsen et al., 2024). Stimuli were individually amplified based on the pure-tone audiogram, and the test was repeated to assess reliability. Participants listened over headphones to sequences of noise segments, in which one segment occasionally contained a modulated (“siren-like”) quality, and indicated when they detected this modulation. The modulation strength was adaptively varied using a threshold-tracking procedure similar to that employed in pure-tone audiometry (Hughson & Westlake, 1944) to determine the minimum contrast at which the participant could reliably detect the target.

HA DIR + NR Feature

The HA used in this study was equipped with a noise management system, as described in Andersen et al. (2021), which applies adaptive beamforming and postfiltering to improve speech intelligibility in challenging listening environments. The definition of the available DIR + NR settings (NPs) is shown in Table 1.

Figure 2 displays the SNR enhancement provided by the different NPs relative to the raw input SNR. Measurements were obtained using the same test setup as that used for the SRT described in section Audiologic measures. SII-weighted output SNR was estimated using the phase-inversion method from Hagerman & Olofsson (2004) across input SNRs ranging from −10 to +10 dB for linear amplification, N3 audiogram (Bisgaard et al., 2010), and custom domes with no vent, fitted to a head-and-torso simulator (Brüel & Kjær type 4128-C HATS) manikin. Across the input SNR range, NPs 2 through 7 provided progressively increasing SNR enhancement with decreasing input SNR. Note that the magnitude of SNR enhancement differences between compared help-level pairs (as defined in section Audiologic measures) was not constant across the −5 to +5 dB range. For example, the difference between NPs 6 and 2, compared in field period (FP) 1, ranged from 5.6 to 6.9 dB, while in FP2, differences were markedly smaller and ranged from 1.9 to 3.9 dB for the NP4 – NP2 comparison and from 2.6 to 3.7 dB for the NP6 – NP4 comparison.

OPEN IN VIEWER

Figure 2.

SNR enhancement relative to the HATS output SNR measured without HA, shown as a function of input SNR for custom domes (no vent) and for linear amplification. Empty-circle data points indicate a 6–10 dB difference between the residual error and the target or masker level, reflecting measurements that are less reliable but still valid. HA = hearing aid; SNR = signal-to-noise ratio.

Study Design

The study was designed for real-world field testing, where participant engagement was a major constraint. To maximize the likelihood of perceptible differences early on, each participant's comparison sequence began with the largest and most meaningful contrast in feature strength, and then progressively moved toward smaller contrasts, refining their choices over time. Although alternative designs could have been methodologically cleaner, their duration and complexity would have made them impractical under these real-world conditions.

Specifically, the field study comprised six 4-week periods in which participants compared two HA programs (P1 and P2) per FP. The study evaluated two key HA features: DIR + NR during the first four periods, and brightness and soft gain (B + SG, gain above 1 kHz) during periods five and six. Each participant followed an individualized comparison trajectory, beginning with large contrasts. For the DIR + NR comparisons, a final preferred NP was determined for each participant by the end of the fourth period. The fifth and sixth periods focused on evaluating preferences across different B + SG settings, with DIR + NR fixed at each participant's final preferred setting. Because the B + SG evaluation addressed a secondary exploratory question, it was positioned at the end of the study to avoid excessive participant burden earlier on.

Program assignments across FPs were as follows:

FP1: Participants compared NP2 and NP6.

Across all FPs, the two programs (P1 and P2) were counterbalanced: half of the participants received the first setting in P1, and the other half received the second setting in P1. P1 was always the starting program in the HA due to the device configuration. The study design is summarized in Figure 3.

OPEN IN VIEWER

Figure 3.

Study design guiding participants through an individualized trajectory of DIR+NR and B+SG comparisons. Experimental design comprising 6 FPs lasting ∼4 weeks; 2 listening programs (P1 and P2) in each FP with decreasing program differences. Programs differed in: DIR+NR strength in FP 1-4 and B+SG in FP 5-6. NPs 2 to 6 were adaptive DIR+NR settings. NP1 indicates the pinna omni with no NR and NP7 the full DIR with max NR. The circles and brackets show the available programs and program comparisons, respectively, for each FP. The blue lines show an example of a participant's choice trajectory. Δ indicates the number of steps separating the two programs. DIR+NR = directionality and noise reduction; NP = noise program; FP = field period.

For each FP, a preferred setting per participant was derived based on a structured interview conducted at the end of each field trial. Participants were asked two questions: (a) “Which setting do you prefer overall?” with response options ranging from “first much better,” “first somewhat better,” “no preference,” “second somewhat better,” and “second much better.” If “no preference” was chosen, participants were asked whether they could hear a difference between the two programs and, if so, to describe it. In this case, the program with the highest average duration of use per day was selected as the preferred program for the specific FP. (b) “How sure are you about this choice?” rated on an 11-point scale ranging from “very unsure” to “very sure.”

During each FP, participants completed self-initiated field report questionnaires whenever relevant listening situations occurred. They were encouraged to switch between programs in these self-chosen situations and completed the questionnaire after listening through both programs, with instructions to provide at least one report per day. German participants received a small monetary incentive per completed survey. Japanese participants received the same switching instructions but were compensated for overall participation rather than per survey.

Reports were submitted either online (via smartphone or computer) or on paper. The German questionnaire was translated into Japanese and back-translated to ensure accuracy. First, participants selected the type of situation from a predefined menu of nine options: conversation with one person, conversation with several people, other people's speech or conversation, music, music via streaming, a broadcast (TV/radio/podcast), a phone conversation, everyday sounds, or other. For each situation, participants rated its importance to them on an 11-point scale (0–10) and the noisiness of the environment on the same 0–10 scale, allowing also for any intermediate values (e.g., 5.5). Participants also reported their program preference using an 11-point scale ranging from −5 (P1 much better) to +5 (P2 much better), with 0 indicating no perceived difference. For each situation, participants answered the following questions: overall support—“Comparing both programs, which program better supported your hearing in this situation?”; sound quality—“Comparing both programs, which program delivered better sound quality in this situation?”; and speech understanding—“Comparing both programs, with which program was it easier to understand what is being said?” (answered only when the situation included speech).

Data Analysis

To address the first research question about which settings users prefer, the final confidence-weighted preferred DIR + NR and B + SG settings selected by each participant were analyzed using descriptive statistics. To convert final preferences to confidence-weighted preference scores, the following steps were taken. Final preferences in every FP were collected on a 5-point scale (0–4; “first much better” to “second much better”), as described in the Study design section. This scale was first centered around zero yielding a symmetric range from −2 to +2, where 0 represents “no preference.” To account for response certainty, each centered preference rating was weighted by the participant's confidence, reported on an 11-point scale (0–10) and normalized to the unit interval [0, 1]. Multiplying preference by normalized confidence shrinks uncertain ratings toward 0 (no preference), while preserving the magnitude of high-confidence ratings toward either end of the scale. The resulting score was then rescaled such that the extremes corresponded to the two conditions tested in each FP.

We also compared SRT benefits (ΔSRT) for the “preferred” and “strong” (NP6) settings, calculated as the difference between each setting and NP1. Differences in SRT benefit were analyzed using a repeated-measures ANOVA with HA condition as a fixed factor.

To address the second research question, whether participants were sensitive to changes in DIR + NR settings relative to minor gain adjustments, we examined situation-specific preference strength, defined as the absolute value of participant ratings. The data were analyzed using a generalized linear mixed-effects model. Sensitivity to program differences was further assessed using end-of-FP interview data, i.e., the proportion of participants reporting no audible difference between programs and their perceived amount of support across FPs.

To investigate the third research question, associations between audiologic factors (aided SRT with DIR + NR off, ACT value, audiogram, degree of acoustic coupling, age and HA experience) and environmental factors (subjective noisiness rating and sound environment) with DIR + NR and B + SG preferences were examined. Final confidence-weighted preferences were analyzed as a continuous outcome using a linear mixed-effects model. Situation-specific preference ratings, originally on a continuous −5 to 5 scale, were converted into a three-level ordinal factor (Low < No difference < High where “No difference” corresponded to values strictly between −1 and 1, and were analyzed with a cumulative link mixed model (CLMM). In addition, we explored whether grouping participants based on their speech-in-noise performance and subjective noisiness ratings would uncover differences in their final DIR + NR and B + SG program preferences, using k-means clustering.

All statistical analyses were conducted in R (Version 4.4.1; R Core Team, 2024). Generalized linear mixed models were fitted using the glmmTMB package (Brooks et al., 2017), and CLMMs using the ordinal package (Christensen, 2023).

Participant Engagement and Data Overview

Table 2 provides an overview of participant attendance, audible difference reports, and field report submissions across the six FPs, with details discussed in the relevant sections below.

OPEN IN VIEWER

Table 2.

Overview of Participant Compliance Across the Six FPs, Including Attendance and Proportion Reporting No Audible Difference Between Programs During End-of-Period Interview, and Number of Participants Submitting at Least One Field Report.

FP	Participants Attending Interview	No Audible Difference Reported (P1 vs P2) During Interview	Participants Submitting ≥1 Field Reports
FP1	123	19 (15.4%)	116
FP2	121	35 (27.3%)	115
FP3	118	36 (31%)	110
FP4	117	40 (34.2%)	109
FP5	114	5 (4.39%)	93
FP6	114	7 (6.14%)	89

FP = field period.

Sound Environments

Figure 4(a) shows the relative occurrences of the diverse listening situations reported. Relative occurrence rates were first calculated for each participant by computing the percentage of that participant's reports assigned to each situation-category. These participant-level percentages were then averaged across participants to obtain the mean and SD for each category. Preference ratings for speech understanding were collected in five speech-related situations (“Conversation with one person,” “Conversation with several people,” “Other people's speech or conversation,” “A broadcast [TV/radio/podcast],” and “A phone conversation”). Overall, 71% of self-report situations included speech. Boxplots in Figure 4(b) further summarize listener-level mean noisiness ratings for each category: the boxes represent the interquartile range (25th to 75th percentiles), the horizontal lines inside the boxes indicate the medians, the whiskers extend to 1.5 times the interquartile range, and points beyond the whiskers are plotted individually as outliers. The mean subjective noisiness ratings, ranged from 2.4 to 5.5, indicating that participants experienced environments varying from relatively quiet to moderately noisy. Importance ratings were strongly negatively skewed (skewness = −1.2), with a 25th percentile of 6.1.

OPEN IN VIEWER

Figure 4.

(a) Relative occurrence of self-reports and (b) subjective noisiness ratings across participants for nine listening situation-categories.

Final DIR + NR and B + SG Preference

Figure 5a shows the final confidence-weighted preferred DIR + NR setting chosen by the 117 participants before proceeding to the B + SG FP testing. There was variability in the preferred DIR + NR setting, with a strong overall preference for the adaptive DIR + NR mode (NP2 to NP6, selected by 87% of participants) compared to the omnidirectional with NR off (NP1 = DIR + NROff) and fully directional with NR max (NP7 = DIR + NRMax) settings. Figure 5b shows the final confidence-weighted preferred B + SG setting chosen by the 114 participants at the end of FP6. Both DIR + NROff and B + SG distributions appeared approximately symmetric, with skewness values of 0.16 and 0.26, respectively.

OPEN IN VIEWER

Figure 5.

Distribution of confidence-weighted (a) DIR+NR and (b) B+SG preference settings at the end of FP4 and FP6, respectively, based on the structured interviews. DIR+NR = directionality and noise reduction. FP = field period.

Differences in SRT benefit between the “preferred” and “strong” (NP6) settings relative to NP1 baseline were analyzed using a repeated-measures ANOVA with HA condition as a fixed factor and participant as a random factor. A total of 114 participants were included; those with missing data for either condition were excluded. The strong setting yielded a significantly higher SRT benefit (ΔSRT = 4.81 dB SNR, SD = 2) compared to the preferred setting (ΔSRT = 3.39 dB SNR, SD = 2) (F(1,113) = 36.34, p < 0.001), as illustrated in Figure 6.

OPEN IN VIEWER

Figure 6.

ΔSRTs measured with the two different HA settings (“preferred” and “strong”). The thick black lines depict the group averages, the shaded areas represent ±1 standard deviation around the mean, and the black dots show the individual results, connected by a gray line for each participant. HA = hearing aid; SRT = speech reception threshold.

User Sensitivity to Changes in DIR + NR Versus B + SG

Situational Preferences

We aggregated all ratings of perceived overall support, sound quality, and speech understanding from participants with available field reports. Histograms (Figure 7) illustrate program comparison ratings across DIR + NR and B + SG FPs, with ratings ranging from −5 (preference for the milder setting) to +5 (preference for the stronger setting) and 0 indicating no perceived difference. Ratings from the full dataset are shown in light gray, while ratings restricted to situations with higher subjective noisiness (noisiness > 5) are shown in dark gray.

OPEN IN VIEWER

Figure 7.

Histograms of the pooled situation-specific preference ratings for each FP (FP1-4: DIR+NR; FP5-6: B+SG). Light gray bars represent all ratings, while darker gray bars represent program ratings from situations in which participants reported subjective noisiness ratings in the upper half of the scale (i.e., >5). Negative values indicate a preference for a milder setting, positive values indicate a preference for a stronger setting, and a value of 0 indicates no perceived difference. For FP 2, 3, 4, and 6, users experienced different setting comparisons defined by their individual choice trajectory. Numbers in parentheses indicate the total number of ratings and participants contributing to each FP. DIR+NR = directionality and noise reduction; FP = field period.

Note that in FP2, FP3, FP4, and FP6, participants experienced different program comparisons depending on their individual choice trajectories. Preference distributions in DIR + NR periods (FP1 to FP4) had sharper peaks at rating = 0. A similar pattern was observed when examining program ratings from situations with higher subjective noisiness.

Situation-specific preference strength, defined as the absolute value of participant ratings (overall support, sound quality, and speech understanding), was analyzed using a generalized linear mixed-effects model with a Tweedie distribution (Tweedie, 1984) and log link, appropriate for positive, positively skewed outcomes including zeros.

Feature type (DIR + NR or B + SG), FP, and test site were included as fixed effects, and participant as a random intercept. Results showed that participants’ preference strength differed significantly by feature type (DIR + NR or B + SG), FP, and test site. Regarding feature type, preference strength was higher for B + SG: participants in FP5 and FP6 showed greater log preference strength compared to the average of the baseline DIR + NR conditions (FP1-FP4) (β = 0.994, SE = 0.033, z = 30.00, p < .001), corresponding to approximately 2.7 times higher preference strength. Across FPs, preference strength decreased (β = –0.167, SE = 0.009, z = –19.03, p < .001), with exp(−0.167) ≈ 0.846 indicating roughly a 15% reduction per FP. Regarding test site, German participants reported higher preference strength (β = 0.496, SE = 0.146, z = 3.41, p < .001), corresponding to ∼64% higher preference strength (exp(0.496) ≈ 1.64) compared to Japanese participants. The random intercept for participants indicated substantial variability among participants (variance = 0.553, SD = 1). Model details are reported in Supplementary Table S1.

Audiologic and Environmental Effects

We examined associations between audiologic (aided SRT with DIR + NR off, ACT value, audiogram, degree of acoustic coupling, age, and HA experience), test site, and environment-related factors (subjective noisiness rating and sound environment) and participants’ preferences for DIR + NR and B + SG.

Participant Clustering

Only participants who submitted field reports for all four DIR + NR FPs were included (n = 102). The elbow method (Thorndike, 1953) was used to determine the optimal number of clusters. The elbow method suggested three clusters based on speech-in-noise performance (aided SRT with DIR + NR off) and subjective noisiness ratings. Clustering was performed using k-means, with both variables scaled prior to analysis. The subsequent k-means analysis yielded cluster sizes of 27 (JP 13, DE 14), 41 (JP 10, DE 31), and 33 (JP 14, DE 19) participants. The quality of clustering was further assessed using the silhouette method (Rousseeuw, 1987). The average silhouette score indicated moderate to good cluster separation (s = 0.38), suggesting reasonably distinct groups despite some overlap. The distributions of the clustering variables are shown in Figure 8a and b; distributions of the remaining variables are presented in Figure 8c–h.

OPEN IN VIEWER

Figure 8.

Clustering of participants based on (a) aided SRT for DIR+NR_Off and (b) mean subjective noisiness ratings. Panels (c–h) show the distributions of final DIR+NR and B+SG preferences and other audiologic variables across the three clusters. The black lines depict the group average, and the shaded areas represent ±1 standard deviation around the mean. DIR+NR = directionality and noise reduction; SRT = speech reception threshold.

The cluster effect was subsequently examined using one-way ANOVA followed by post-hoc Tukey tests. A one-way ANOVA revealed a significant effect of cluster on aided SRT for DIR + NR_Off (F(2, 98) = 71.88, p < 0.001) and mean subjective noisiness (F(2, 98) = 66.71, p < 0.001). For aided SRT for DIR + NR_Off, Tukey's HSD post-hoc tests indicated that Cluster 1 had significantly higher SRTs than Clusters 2 and 3, which did not differ from each other. For the mean subjective noisiness, post-hoc tests showed that all clusters differed significantly from each other, with Cluster 2 reporting the highest and Cluster 3 the lowest mean subjective noisiness.

After cluster definition, we examined the distributions of final DIR + NR and B + SG preferences and other audiologic variables (Figure 8c–h). A one-way ANOVA on final weighted DIR + NR and B + SG preference revealed no significant effect of cluster for either condition: DIR + NR, F(2, 98) = 0.29, p = .75, and B + SG, F(2, 96) = 0.39, p = .68. Effects were observed for several audiologic measures, with Cluster 1 characterized by higher ACT values, older age, higher audiogram thresholds, and lower REOIG compared to Cluster 2 (all p < 0.05). Post-hoc Tukey tests indicated that ACT and audiogram also differed significantly between Cluster 1 and Cluster 3 (all p < 0.001), whereas differences in REOIG and age were only significant between Cluster 1 and Cluster 2.

Participant Engagement

Participants showed high overall engagement with substantial inter-individual variability. Reporting frequency was lower than instructed, with relatively few reports from highly noisy situations (section Field report completion). HA use was high (section HA use). Participation remained largely stable, while report submissions declined over time (section Participant compliance). Participants reported across diverse listening environments (section Sound environments) in line with other EMA-studies (Borschke et al., 2024; Christensen et al., 2024a). Field ratings showed strong agreement with end-of-period preferences (section Consistency of participant program preference ratings), likely reflecting that retrospective evaluations were dominated by the programs participants used most frequently, demonstrating the reliability of the field ratings.

Final DIR + NR and B + SG Preference

Participants largely preferred adaptive DIR + NR settings (NPs2 to 6, 87%), with preference distribution resembling a normal distribution with modest DIR + NR in the center (section Final DIR + NR and B + SG preference). However, speech-in-noise testing showed that DIR + NR preferences did not maximize SRT benefit, as the DIR + NR_Strong setting provided significantly greater SRT improvement than the participants’ preferred settings. This indicates that substantial additional speech-in-noise benefit remained unused when relying solely on subjective preference. In contrast, preferences for B + SG settings were more equally distributed.

User Sensitivity to Changes in DIR + NR and B + SG

Situation-specific ratings revealed weaker preferences during the DIR + NR phases (FP1 to FP4), compared with preferences during the B + SG phases (FP5 and FP6). The weakness of preferences was more pronounced for the Japanese participants (section Situational preferences). Several factors may explain this. One possibility is that environmental differences played a role: 10 (27%) Japanese participants were classified in the highest subjective noise cluster (Cluster 2), compared with 31 (48%) of German participants, suggesting that the two groups may have been exposed to different acoustic conditions during testing, which could influence the strength of their preferences. Alternatively, cultural differences in scale-use behavior may also contribute, for example if one group tends to use rating scales more conservatively than the other.

Overall program preferences followed the same pattern, with reports of “no audible difference” being more common during DIR + NR comparisons (mean ∼27%, increasing from FP1 to FP4) but rare during B + SG (∼5%). Despite most participants being satisfied with the amount of help provided by DIR + NR, a substantial proportion of participants indicated that the amount of DIR + NR support provided was “not enough” (∼29%) (section Program perception). Overall, after extensive testing using an individualized comparison trajectory for each participant, DIR + NR preferences were weaker and less consistent than the B + SG preferences.

The DIR + NR feature was designed to provide nearly 7 dB of SNR improvement between NP2 and NP6, and for most help-level contrasts (except in FP4), the differences between high and low NPs reached 3 dB or more (see Figure 2). These technical contrasts aligned well with the recommended minimum benefit needed for reliable discrimination of NR schemes (McShefferty et al., 2015). However, despite these substantial laboratory-measured contrasts, the corresponding differences appear to have been less prominent in real-life use.

This observation is consistent with previous findings showing that laboratory-measured benefits of DIR + NR features often do not translate to clear real-world listening advantages. Wu et al. (2019) reported that premium DIR + NR features produced measurable laboratory improvements, but did not yield clear real-world benefits. Similarly, Gnewikow et al. (2009) found substantial objective speech-in-noise advantages for directional processing across hearing loss groups, yet subjective ratings did not reflect a clear preference for directional amplification in daily life.

One possible explanation for the lack of real-life benefit is that the effective SNR differences in daily life were reduced relative to the measured ones. Several factors likely contributed to this attenuation. First, the measured SNR differences in Figure 2 were obtained with closed earmolds, whereas participants in the present study had a large variety of closedness of acoustic coupling ranging from completely occluded to open bass domes. More occluded fittings are related to larger SRT benefit from DIR + NR (Jürgens et al., 2025). Therefore, the openly fitted HA users may have experienced smaller SNR-contrasts between two DIR + NR settings due to the unprocessed sound entering the ear canal and partially obliterating the contrast. Thus, the type of acoustic coupling is likely a key factor influencing strength and consistency of DIR + NR preferences. Overall or situational preference for specific DIR + NR settings did not correlate with the closedness of acoustic coupling (i.e., the REOIG).

Second, everyday life scenarios reported by HA users may not have the same spatial characteristics as the setting used for the technical SNR-improvement measurements in Figure 2, which is favorable for DIR processing. Therefore, participants may have experienced also in their HA amplified sounds weaker SNR improvements than suggested by the technical evaluations here. Therefore, using more ecologically valid scenarios with more diffuse background noise for technical evaluations would likely reduce the measured benefit and further help bridge the gap between laboratory measurements and real-life performance (Best et al., 2015).

Third, the benefit of DIR + NR is smaller at the input SNRs most commonly experienced by HA users. As shown in Figure 2, the technical SNR improvement decreased as input SNR increased. Since HA users tend to seek out favorable listening situations with average real-world SNRs around +5 dB (Smeds et al., 2015; Wu et al., 2019; Borschke et al., 2024), the effective benefit from DIR + NR was likely reduced in the situations they encountered. Also, the SNR differences between high and low NPs were not constant across input SNRs, as noted in section HA DIR + NR feature, meaning that the perceptual contrast between NMSLNPs would vary across environments for each user.

Beyond these reductions in effective SNR, the SNR differences may simply not have been large enough to elicit clear perceptual distinctions. McShefferty et al. (2016) showed that the SNR adjustments required to create meaningful perceptual differences are often substantially larger than those needed for basic discrimination under controlled conditions.

Finally, attentional factors may also weaken DIR + NR preferences. Listeners naturally focus on speech, making them less sensitive to reductions in background noise when the speech signal is largely preserved. This attentional bias may reduce the perceived benefit of DIR + NR, particularly in everyday listening situations where the target speech remains intelligible despite background noise. In contrast, gain changes affect both, speech and noise and lead to an overall more dull or sharp sound percept, which may more easily trigger higher preference strength.

Due to all these factors, participants may have been exposed to smaller-than-expected perceptual contrasts that, depending on the situation, often may have fallen below the discrimination threshold, limiting the strength and consistency of comparisons across NPs. This variability underscores the challenges of evaluating settings in commercial HAs in real life.

Relation of Participants’ Preferences to Audiologic or Environmental Factors

Across analyses for both overall (section Final preferences) and situation-specific preferences (section Situational preferences), none of these factors reliably predicted DIR + NR preferences. Only subjective noisiness ratings showed an effect, with higher ratings associated with a small but significant increase in selecting the stronger DIR + NR setting (β = 0.087, p < .001), consistent with previous findings on DIR processing (Neher, 2014). In contrast, higher noise ratings were associated with a slight decrease in selecting the stronger B + SG setting (β = –0.080, p < .001), likely because higher B + SG levels are perceived as louder and therefore noisier. We did not find a significant effect of test site. However, we cannot fully exclude that differences in HA-fitting methods between the two test sites, specifically that Japanese participants were prescribed 5–10 dB more gain and were fitted with fully closed earmolds, may have influenced participants’ preferences.

The non-existing correlation of age with overall preference for DIR + NR strength in the present study is in line with Wu (2010), who found that preference measured in the laboratory was not age-related. In contrast, Wu (2010) found older age to be significantly correlated with lower preference for DIR in field data, which the present study did not find. This may be due to Wu (2010) using the highest possible contrast only (omnidirectional versus fully directional), whereas the present study used individual trajectories of combined DIR + NR choices and much finer contrasts. The non-existing correlation of BPTA4 with DIR + NR strength in the present study is in line with Neher (2014), who disentangled DIR and NR processing and looked for relations to hearing loss and executive function. Note that when examining only NR (excluding DIR), Neher (2014) found larger PTA to be associated with stronger preference to strong NR. Disentangling DIR + NR was not a goal of the present study; therefore, a direct comparison with their finding is not possible. Future studies could therefore investigate if this preference for strong NR in persons with strong hearing-impairment persists also with modern DNN-based NR-algorithms when excluding DIR.

We found high correlations between perceived speech understanding and sound quality, indicating that improvements in speech understanding did not come at the expense of sound quality (section Consistency of participant program preference ratings). This indicates that, for the specific DNN-based NR implementation evaluated in this study, the speech–quality trade-off is rather limited. This is consistent with the high listener satisfaction reported for the same DNN-based NR algorithm across different levels of background noises (Christensen et al., 2024b). A plausible explanation is that the specific DNN-based NR algorithm used here remains active at all times rather than switching on and off in a binary manner. We therefore speculate that this always-on, continuously adapting mechanism helps maintain sound quality by avoiding abrupt or noticeable changes, contributing to the minimal disruption.

To explore factors that might explain differences in participants’ preferences, we clustered individuals based on their aided SRTs with (DIR + NR)_Off and their subjective noisiness ratings. The resulting clusters differed clearly in both hearing ability and experienced environmental noisiness but showed no differences in final DIR + NR preferences (section Participant clustering). Additionally, the group with the poorest speech-in-noise performance also exhibited higher ACT scores, elevated audiogram thresholds, lower REOIG (more closed fittings), and older age, factors previously identified as predictors of speech-in-noise performance (Zaar et al., 2024; Jürgens et al., 2025).

Listener clustering (section Participant clustering) also showed that participants with poorer speech-in-noise performance, who are likely to benefit the most from DIR + NR, perceived less noise in daily life, consistent with Borschke et al. (2024). This means that for these participants the NR system may not have been sufficiently activated, which is a possible confound within the evaluation of DIR + NR preferences that needs to be acknowledged.

Practical Implications for Fitting DIR + NR Algorithms

While this study was designed to strongly encourage participants to choose between DIR + NR settings over an extended period, preference strength was modest in FP1 and weakened with smaller DIR + NR contrasts up to FP4. In turn, this means that there is a relatively broad acceptance range of different DIR + NR settings. This indicates that clinicians can confidently guide patients toward DIR + NR settings that optimize speech-in-noise performance. However, this approach contrasts with user-driven strategies (Reinten et al., 2023; Johnson and Healy, 2024), with the latter indicating the existence of an SNR “sweet spot,” outside of which excessive NR may reduce environmental sound perception. Therefore, strong DIR + NR settings should not be applied uniformly to all users, and caution is warranted. We propose that, while users’ listening preferences remain important, the broad acceptance of DIR + NR settings supports using speech-in-noise optimization as the primary guide for initial fittings.

Study Limitations

Several limitations should be acknowledged when interpreting the results of this study, besides what has been mentioned above. First, participants had limited exposure to P2 because the HA automatically defaulted to P1 whenever they were switched off and then back on, which may have biased preference measures in favor of P1. Although this does not mean that there is an overall bias in the data across all participants, because assignment of DIR + NR or B + SG processing to P1 and P2 was random, it means that individual participants were exposed to one of the settings more frequently than to the other. Future studies should therefore consider randomizing the startup program both across participants (as done in the present study) and across days to control for potential order effects. Second, the lack of objective noise level measurements limits our understanding of how noise management functionality interacts with subjective noisiness reported by the participants. Thus, in future studies, objectively measured noise level collected using datalogging within the HAs should therefore be compared to subjective ratings. Third, part of the study coincided with the COVID-19 pandemic, reducing participants’ opportunities to experience complex listening situations. Fourth, participants submitted relatively few reports in noisy situations (i.e., noisiness ratings >5), resulting in sparse data precisely when NR is most relevant. Fifth, retrospective reporting may have introduced memory bias, as participants could submit reports at any time without a defined window following the listening event. Finally, the long testing period over half a year showed that the main reduction in engagement was driven by a drop in field report submissions rather than overall participation. Alternative methods should be explored to reduce the reporting burden, e.g., using microinteraction EMA, as demonstrated by Le et al. (2024) and Ponnada et al. (2025).