Amplification self-adjustment: controls and repeatability

Abstract

Objectives.

This study was a continuation of work on an explore-and-select approach to the self-adjustment of amplification. Goals were to determine i) the effect of changing the number of adjustment controls from three to two, ii) the effect of changing the initial adjustment from overall output to high frequency output, iii) individual repeatability, iv) the effect on phoneme recognition of increasing and decreasing overall output relative to the starting and adjusted conditions, and v) listener reactions to, and opinions of, the self-adjustment procedure.

Design.

Twenty-two adults with hearing loss, ten of whom were hearing-aid users, adjusted level and spectrum of connected speech to preference, using three configurations of number and order of adjustment parameters. The three adjustments were replicated to give a total of six. Presentation was monaural, in quiet, using the ear with the better threshold at 2 kHz. The starting condition was a generic prescription for a typical mild-to-moderate hearing loss. Real-ear output spectra were measured for the six self-adjustments, the generic starting condition, and the individual NAL-NL2 prescriptions for speech at 65 dB SPL. Monaural phoneme recognition in monosyllables was assessed, in quiet, at levels of −14, −7, 0 and +7 dB relative to both the starting and the self-adjusted conditions. Participants completed a questionnaire and their comments on each question were transcribed.

Results.

Changing the number of listener controls from three to two reduced mean adjustment time by around 50% but had negligible effect on group-mean output response. Starting adjustment with high-frequency output rather than overall output resulted in a 2 to 3 dB reduction of group-mean self-adjusted output below 1 kHz. Individual self-adjustments were within ±5 dB of NAL-NL2 prescription (for a 65 dB SPL speech input) for two thirds of the participants in the high-frequencies and for just over half in the low-frequencies. In six self-adjustments, individuals self-adjusted, on average, to within ±4 dB of their own mean in both high and low frequencies. There was no evidence that these findings differed for hearing-aid users and non-users. Changes of overall output by ±7 dB after self-adjustment did not significantly affect group mean phoneme recognition. Preference for number and order of self-adjustment differed among participants, as did opinions on self-fitting of hearing aids.

Conclusions.

These findings support the conclusion that, for many adults with hearing loss, an explore-and-select procedure for self-adjustment of amplification leads to output values that are repeatable within a few dB, are relatively immune to the number and order of adjustment parameters, and place the average listener well along the plateau of a phoneme-recognition vs. amplitude function.

Introduction

The study reported here deals with self-adjustment of amplification by adults with hearing loss. It describes continuing work on the understanding of, and optimization of, an explore-and-select approach to self-adjustment of hearing-aid output and spectrum.

The concept of hearing-aid self-adjustment is by no means new. Indeed, a volume control was incorporated into electronic hearing aids when first developed at the beginning of the 20^th century. As hearing aids evolved, digital control of preset analog options offered listeners limited control of spectral shape. The arrival of digital hearing aids enabled detailed multi-band control of frequency-dependent gain, amplitude-dependent gain (compression), maximum output, together with advanced processing features such as directionality, feedback cancelation, noise management, and frequency shifting. At first, this level of control was provided in commercial fitting software and remained under the control of the audiologist or hearing-aid dispenser. An effort to make hearing health-care more accessible, however, has prompted increased attention to user self-fitting and self-adjustment (Convery et al., 2011, 217; Dillon & Keidser, 2011; Donahue et al., 2010; Keidser & Convery, 2016). The US government’s Over-the-Counter Hearing Aid Act of 2017 brought the issue to the fore because purchase without involvement of a professional raises the possibility of serious mismatch between the amplification characteristics of the aid and the auditory needs of the user.

Accepted clinical practice in hearing-aid fitting currently incorporates initial adjustment to a threshold-based prescription followed by fine-tuning (Anderson et al., 2018; ASHA, 1998; Valente et al., 2007). During fine-tuning, the audiologist explores changes from the prescription in response to verbal feedback and selects the client’s preferred response. Adjustments can include any of the amplification parameters mentioned above. This audiologist-driven, explore-and-select, procedure may be repeated after a period of experience with the aid in actual use. At the time of writing, however, several manufacturers have created smart-phone applications that enable limited post-fitting self-adjustment in the field – possibly reducing the need for follow-up appointments.

Complete self-fitting by the client, without professional involvement, represents a major departure from current practice and raises several questions. Some questions concern users’ ability to assemble, insert, and operate the aid – studied at length by Convery and colleagues (Convery et al., 2011; Convery et al., 2019; Convery et al., 2017). Beyond this are questions about client’s ability to evaluate and adjust hearing-aid performance. For example, how should a starting condition be set from which the listener can best explore and select changes? What should the user be listening to? Which of the numerous amplification parameters should be adjusted? How many controls should the user have? How should a small number of user controls be mapped onto the much larger number of amplification parameters? Who is a good candidate for self-adjustment?

Many of these last questions have been addressed in research studies. In the early days of the development of digital hearing aids, Punch and Robb (1992) described a technique involving exploration of, and selection from, a 5*5 matrix of preprocessed sound files with different balances of high- and low-frequency content. The technique was later adapted for a study in which individual self-adjustments of a wearable master hearing aid in the field were compared with, and found to be different from, those obtained in the laboratory (Punch et al., 1994).

Some variations of the explore-and-select procedure for self-adjustment have involved complete or partial automation. Examples include selection from hearing aid options with different pre-programmed characteristics (Davis et al., 2016; Humes et al., 2017; Urbanski et al., 2020); adaptive paired comparison (Kuk & Lau, 1995; Neuman et al., 1987; Preminger et al., 2000); and machine learning (Jensen et al., 2019). Most of the research on self-adjustment, however, has given the listener full control of both exploration and selection, with or without researcher supervision.

Dreschler et al. (2008) explored many of the options involved in creating software and procedures for the fine-tuning stage of self-adjustment. Issues of ear, earmold, and dome acoustics were eliminated by use of amplified sound-field presentation of pre-recorded videos. These videos offered three types of speech input and three of non-speech. Four control configurations were explored for adjustment of gain and spectrum. Individual starting response was prescribed for a 65 dB SPL input using the NAL-RP approach (Byrne & Dillon, 1986). Wide-band 2:1 compression was applied to inputs above 65 dB SPL. Most participants were able to complete and replicate self-adjustment of gain and frequency response speedily and efficiently regardless of hearing-aid experience. Group-mean self-selected gains and spectral slopes were somewhat below prescription and were significantly affected by the type of input and by deliberate deviation from the prescribed starting response. There was no evidence that the type of control affected final self-adjustment, but participants expressed preference for two configurations: a) two-parameter control of overall gain and spectral slope and b) three-parameter control of overall gain, bass, and treble. In discussion, the authors introduced the concept of a range, or zone, of acceptable amplification characteristics for a given input, rather than a single optimum. The existence of such a range would account both for within-participant variations of repeatability and for the influence of starting response on final selection.

Nelson et al. (2018) studied self-adjustment with a two-parameter explore-and-select procedure built into the Ear Machine© (http://www.earmachine.com). The two adjustment parameters are labelled “loudness” and “fine tuning”. The latter adjusts spectral tilt. This is not, however, a simple combination of overall output and spectrum. It uses a system of gain-dependent changes of spectrum and compression ratio that is based on thresholds and prescriptions from a large database of audiograms (Ciletti & Flamme, 2008). Self-adjustments began with individual prescriptions using the National Acoustics Laboratories second Non-Linear approach (NAL-NL2) (Keidser et al., 2012). A principal finding was a wide range of mean deviations from the NAL-NL2 starting response but with negligible effects on mean word-in-sentence recognition score. Nelson et al., also found large effects of noise on self-adjustment but only for signal-to-noise ratios of 0 dB or less. In further analysis of the Nelson et al., (2018) data, Perry et al. (2019), examined many possible predictors of deviations from the NAL-NL2 prescription. They found evidence of effects of hearing-aid experience and degree of hearing-loss but not of age, sex, noise tolerance, or time taken to adjust.

The self-adjustment approach used in the present study initially incorporated the preferred three-parameter approach of Dreschler et al., (2008). Listeners were given control of overall output, high-frequency boost, and low-frequency cut. The process was incorporated into a self-adjustment software package referred to as “Goldilocks” and used in a first study (Boothroyd & Mackersie, 2017; Mackersie et al., 2019). In that study, the spectral-shape component involved exploration of, and selection from, a matrix of pre-processed frequency responses, as in Punch and Robb (1992). There were five levels of high-frequency boost and five levels of low-frequency cut, both in 5 dB steps. Adjustment of overall output was in 3.6 dB steps. Starting response was not based on individual prescriptions but on a generic NAL-NL2 prescription for a typical sloping mild-to-moderate hearing loss. The self-adjustment software tracked, timed, and logged the three adjustments. As others had found, most participants with hearing loss were able to complete self-adjustment speedily and repeatably, regardless of hearing-aid experience. Twenty of the 26 participants (77%) did so without assistance – they simply followed on-screen instructions. In that first experiment, participants were instructed to adjust to preference while listening to sentences spoken by a woman in quiet. No guidance was given as to the relative importance of comfort, quality, or intelligibility. Participants were, however, given a sentence-level speech-perception test after an initial self-adjustment. It was hypothesized that this experience would bias listeners towards intelligibility when adjusting a second time. There was, in fact, evidence of a significant group-mean improvement of Speech-Intelligibility Index between the first and second self-adjustment, but only by experienced hearing-aid users. No replication effect was observed in listeners without hearing-aid experience, and the difference between aid-users and non-users was not present at the second replication.

A second laboratory study, using the Goldilocks approach, employed real-time processing of ear-level microphone input (Mackersie et al., 2020). Processing was accomplished by the UCSD open-source speech processing platform (Garudadri et al., 2017). All participants were experienced hearing-aid users. This time, however, there was no evidence of a significant group-mean change of self-adjustment after the formal speech-perception test. A possible reason for failure to replicate was the ear-level microphone, which provided listening experience from both researcher instructions and the participant’s own speech. There was weak evidence of convergence towards NAL-NL2 prescription at the second self-adjustment. Participants whose first adjustment was below prescription tended to increase output at the second. Participants whose first adjustment was above prescription tended to reduce output at the second. This finding suggested a benefit from allowing more than one self-adjustment – a conclusion also reached by Dreschler et al. (2008).

The Mackersie et al., 2020 study also assessed the effect of noise (at a signal-to-noise ratio of 6 dB) on self-adjustment. No evidence of such an effect was found. As expected, however, the noise did significantly affect speech perception. Less expected was that it increased the reporting of intelligibility as the primary criterion for self-adjustment.

A feature of the Goldilocks self-adjustment software is that it allows presentation of each control in isolation before showing all together. The intent is to help participants learn the relationship between adjustment of each control and the resulting changes of sensation. It is hypothesized that this will help avoid confusion when all controls are shown simultaneously. In the two studies just described, the sequence of adjustments during this learning process was volume, treble, bass (i.e., overall output, high-frequency boost, and low-frequency cut), after which all three were shown simultaneously.

A first goal of the present study was to compare this three-parameter control configuration with two-parameter control of overall level and spectral slope. A second goal was to determine the effect, if any, of changing the sequence by allowing adjustment of high-frequency output first. Additional goals were to determine within- and between-participant repeatability of self-adjustments and to measure phoneme recognition for levels at, below, and above self-adjustment. Finally, we sought listener reactions to, and opinions about, this self-adjustment process and self-fitting in general.

In summary, the present study was designed to address the following questions:

1. How are the level and spectrum of self-adjusted output affected by changing from 3-parameter control (Volume, Treble, Bass) to 2-parameter control (Volume, Spectral Slope)?

2. How are the level and spectrum of 3-parameter self-adjusted output affected by changing the first adjustment from overall amplitude to high-frequency amplitude?

3. How do group-mean and individual self-adjustments compare with both a generic starting response and the NAL-NL 2 prescribed starting response?

4. How repeatable are group-mean and individual self-adjustments of high-and low-frequency output?

5. How is phoneme recognition in consonant-vowel-consonant words affected by increase or decrease of amplitude relative to both the generic starting response and the self-adjusted responses?

6. What are participants’ opinions about the explore-and-select approach to self-adjustment and self-fitting?

Materials and Methods

Participants

Twenty-two adults with mild-to moderate hearing loss participated. Ten hearing-aid users (four men, six women) were recruited from the San Diego State University audiology clinic population. Twelve non-users (seven men, five women) were recruited via advertisement in local newspapers. Inclusion criteria were i) a sensorineural hearing loss of at least 40 dB HL at 2 kHz and above, ii) English fluency, and iii) a score on the Montreal Cognitive Assessment (MoCA) of 21 or higher (Nasreddine et al., 2005). Based on the repeatability data from the Mackersie et al., (2020) study, a sample size of 20 was estimated to have 84% power of detecting a group-mean output difference of 2 dB, in a single half-octave frequency band, with a confidence level of .05.

Note that the participant sample included examples of both hearing-aid users and non-users. There was no attempt, however, to balance the two groups in terms of hearing loss, age, or sex.

One ear was tested during this study. This was the ear with the better threshold at 2 kHz. If there was no difference, the test ear was selected at random. Table 1 shows background information for the participant sample. Figure 1 shows median, ranges and quartiles of the test-ear pure-tone thresholds. Also shown in Figure 1 is the generic hearing loss used to prescribe the starting response for all participants. This response was some 10 dB less severe than the median loss for this sample at frequencies above 1000 Hz.

Table 1.

Background information for 22 participants

	Aid	Aid	MoCa	Age			kHz						4-freq
ID	use	yrs	score	(yrs)	Sex	Ear	0.25	0.5	1	2	4	8	Avg.
1	NO		30	76	F	R	25	20	30	45	55	65	38
2	NO		26	74	M	L	10	15	40	55	65	60	44
3	NO		26	66	M	R	15	10	20	40	65	55	34
4	NO		27	64	F	L	30	40	35	45	60	75	45
5	NO		28	69	M	L	50	50	55	60	75	65	60
7	NO		27	71	M	R	15	15	25	50	50	55	35
12	NO		22	63	M	L	35	35	40	55	60	55	48
17	NO		28	58	M	L	20	20	30	60	65	50	44
19	NO		29	53	F	R	45	45	45	60	65	70	54
20	NO		29	80	M	R	20	35	30	45	60	70	43
21	NO		26	93	F	L	25	25	45	50	75	100	49
22	NO		30	46	F	R	55	60	60	60	40	45	55
6	YES	0.5	28	69	M	L	15	10	45	60	65	80	45
8	YES	8	26	68	F	L	40	40	50	65	90	105	61
9	YES	0.5	28	68	M	L	15	20	30	45	45	70	35
10	YES	2	21	73	F	R	20	15	20	55	50	50	35
11	YES	2	28	72	F	L	15	10	25	60	80	75	44
13	YES	1	27	81	M	R	35	40	40	50	55	60	46
14	YES	4	28	85	F	L	35	35	35	40	55	55	41
15	YES	10	23	83	M	R	20	20	20	60	60	60	40
16	YES	6	28	89	F	L	65	60	55	60	60	75	59
18	YES	38	22	91	F	R	70	70	65	65	65	70	66
	M	7	26	78			33	32	39	56	63	70	47
	SD	11	3	9			20	21	15	8	14	16	11

Notes. MoCa = Montreal Cognitive assessment scale.

4-freq Avg. = average of thresholds at 0.5, 1, 2, and 4 kHz.

Figure 1.

Distribution of air-conduction thresholds.

Speech materials

1. During self-adjustment, participants heard concatenated City University of New York (CUNY) sentences prerecorded by a woman (Boothroyd et al., 1988). In these materials each sentence is meaningful but sentences within in a set are unrelated. In other words, there is sentence context but no narrative context. The same sentence materials were also used for assessment of real-ear output spectra, but not for intelligibility assessment.

2. The intelligibility outcome measure was phoneme recognition in isolated Consonant-Vowel-Consonant words. This measure used the AB isophonemic word lists, recorded by the same woman talker (Boothroyd, 2008). Each of the 20 lists consists of 10 consonant-vowel-consonant words containing one example each of same 10 vowels and 20 consonants.

Preprocessing

The sentence and word recordings were digitized at a resolution of 16 bits and a sampling rate of 22050 Hz. They were then subjected to instantaneous 1.5:1 wide-band amplitude compression with effective attack and release times of 2.5 msec. Spectral processing used five bands with nominal center frequencies at octave intervals from 500 to 8000 Hz. Infinite Impulse Response (IRR) low and high-pass filters were used for bands 1 and 5, respectively. IRR band-pass filters were used for bands 2 through 4. A matrix of 49 spectra was created for each sentence or word. This matrix provided seven 3 dB steps of high-frequency level and seven 3 dB steps of low-frequency level. The pivot point for change of spectral slope was 800 Hz. Processing was accomplished with DaDisp 6.7 software (DSP Development Corporation, 320 Nevada Street, Suite 301, Newton, MA 02460). Figure 2 shows the range of the resulting gain curves. The heavy broken line shows the gain curve at which self-adjustment began. With the sentence material used during self-adjustment, this curve produced an output spectrum close to the NAL-NL2 starting prescription for the generic mild-to-moderate sensorineural hearing loss shown by the circles in Figure 1.

Figure 2.

Gain curves showing seven levels of high-frequency boost and seven levels of low-frequency cut.

Self-fitting software

The Goldilocks explore-and-select software used in Boothroyd and Mackersie (2017) and Mackersie et al., (2019) was adapted for the present study. As before, the participant could adjust three parameters labelled “Loudness” (overall amplitude, also referred to as Volume), “Crispness” (high-frequency boost) and “Fullness” (low-frequency cut). The number and size of the steps of overall-amplitude change were the same as in the original version of Goldilocks (12 levels in steps of 3.6 dB). The number of levels of both high-frequency boost and low-frequency cut, however, was increased from 5 to 7 and the step size was reduced from 5 to 3 dB. As before, the parameter controls were first shown, and adjusted, one at a time. The goal was to help the participant learn the changes of sensation associated with each control. All controls were then shown simultaneously for final adjustment, as illustrated in Figure 3.

Figure 3.

The three controls for final self-adjustment, shown simultaneously.

For the present study, two modules were added to the Goldilocks self-fitting software. One enabled presentation and researcher-scoring of the phoneme-recognition test. The other was a self-administered questionnaire.

Two self-adjustment methods were also added, providing a total of three:

1. Three-parameter, volume-first (VHL): i) Volume, ii) High frequencies, iii) Volume readjust if needed, iv) Low frequencies, v) all three for final adjustment. This was the sequence used in the previous two studies.

2. Two-parameter, volume-first (VH): i) Volume, ii) High frequencies, iii) both for final adjustment. In this protocol, any high-frequency change was accompanied by an equal and opposite low-frequency change. In other words, the “Crispness” control became one of overall spectral tilt and the “Fullness” control was hidden from the participant.

3. Three-parameter, crispness-first (HVL): i) High frequencies, ii) Volume, iii) High-frequency readjust if needed, iv) Low frequencies, v) all three for final adjustment.

Questionnaire

Five questions, with five response alternatives, explored the participant’s opinions on the fitting process:

1. What had the MOST influence on your sound adjustments?

2. How clear were the instructions?

3. How do you feel about the number of things to adjust?

4. How do you feel about the adjustment to start with?

5. Who do you think should be responsible for hearing-aid fitting?

Equipment

1. The stimuli and the Goldilocks self-fitting software were presented from a Microsoft surface Pro 4 computer with docking station. Participants listened monaurally to the output of an HF5 earphone (Etymotic Research Inc. 61 Martin Lane, Elk Grove Village, IL 60007). Flexible triple-flange silicone domes were used to couple receiver to ear.

2. Real-ear output spectra, for the CUNY sentence material, were measured with a Verifit2 hearing-aid analyzer (AudioScan, 20 Ludwig St. Dorchester, Ontario N0L 1G4, Canada).

Procedure

Test sessions

Pure-tone threshold and MoCA testing took place in a preliminary session to establish candidacy. Participants were briefed on the experiment and gave informed consent. Experimental testing, described below, was completed in a second session, lasting 90-minutes.

Instructions

After insertion of the earphone in the test ear, participants read the following on-screen instructions:

“You will hear a woman talking. Imagine you will be listening to her on radio or TV for 20 minutes or so. 1) Use “More” to increase the sound feature until it is too much. 2) Use “Less” to decrease the sound feature until it is too little. 3) Continue to adjust “More” and “Less” until you find a setting that is best for you.4) Hit OK when you are satisfied. Any questions? Hit START when you are ready to begin listening.”

In every-day use of hearing aids, each listener must find his or her own compromise among adjustment criteria such as loudness, comfort, effort, intelligibility, and sound-quality. For this reason, the only criterion offered here was “best for you”.

Self-adjustments

While listening to the concatenated sentences, participants adjusted output and spectral shape using each of the three protocols in turn. Each self-adjustment began with the generic starting response. The protocol sequence for odd-numbered participants was 3-parameter crispness-first (HVL), 3-parameter-loudness-first (VHL), 2-parameter-loudness-first (VH). The sequence for even-numbered participants was VH, VHL, HVL. The original Goldilocks VHL protocol was always in the second position while the two experimental protocols alternated in first and third position. After the initial round of three self-adjustments, participants were given a break before repeating the process, using the same protocol sequence, for a total of six self-adjustments. No speech-perception testing took place between the first and second set of self-adjustments. In the absence of sound-field presentation and microphone input, participants had no additional experience of listening with their first three self-adjusted responses.

Questionnaire

Following the second round of self-adjustments, the adjustment questionnaire was administered. After each question, the participant was invited to comment, and any comments were noted by a research assistant for later exploration of themes.

Speech perception

Phoneme recognition was assessed using the second self-adjusted responses beginning with volume (VH and VHL), and with amplitude changes of +7.3, −7.3, and −14.5 dB. The process was repeated for the generic starting response.

Real-ear testing

Finally, real-ear output spectra were measured for the six self-adjustments, and the generic starting response. The input signal was the same CUNY sentence material heard during self-adjustment but with gaps between sentences removed. For comparison with these empirical measures, individual NAL-NL2 prescribed real-ear output targets, for a 65 dB SPL speech input, were obtained from the Verifit. There was no correction for sex or hearing-aid experience.

Results

Group-mean real-ear responses

An initial analysis failed to show significant effects of replication, either as a main effect or in interaction with adjustment protocol or frequency. Data were, therefore, collapsed across the two replications for examination of the effects of order and sequence of adjustment parameter.

Data points in the left panel of Figure 4 illustrate the effect of changing from three (VHL) to two adjustment parameters (VH). This effect was examined in a repeated-measures analysis of variance with method at two levels and frequency at ten levels. Greenhouse-Geisser corrections for non-sphericity reduced degree of freedom for frequency from 9 to 3.5 and for the method x frequency interaction from 9 to 1.5. The average outputs for the three- and two-parameter methods were 64.4 and 63.4 dB SPL, respectively. The difference of 1 dB approached, but did not reach, the .05 level of significance (F(1,21) = 3.9, p = .062, ⴄ²_p = 0.16). Differences at individual frequencies ranged from 0.8 to 1.9 dB, but there was no evidence of significant interaction (F(1.8,37.4) = 0.56, p = .57, ⴄ²_p = 0.03). After Greenhouse-Geisser correction, the main effect of frequency remained large and highly significant (F(3.5,73.8) = 104, p <.0005, ⴄ²_p = .83). These data do not, however, provide evidence that changing from three to two adjustment parameters affected group-mean output-level or spectral shape.

Figure 4.

Group-mean real-ear output responses: starting, NAL-NL2 target, and self-adjusted.

Data points in the right panel of Figure 4 illustrate the effect of changing the three-parameter sequence from Volume-first (VHL) to Crispness-first (HVL). The average outputs were 64.4 and 63.1 for the two methods. In repeated-measures analysis of variance, the difference of 1.2 dB just failed to reach the .05 level of significance (F(1,21) = 4.1, p = .056, ⴄ²_p = .16). Even after Greenhouse-Geisser correction, however, there was a significant interaction between method and frequency (F(1.9,40.2) = 4.8, p = .015, ⴄ²_p = .18). Neuman-Keuls post-hoc analysis showed highly significant differences from 250 to 1000 Hz, ranging from 2.2 to 2.8 dB (p < .0005). But, at frequencies above 1kHz, the differences ranged from 0.1 to 0.8 dB and none reached the .05 level of significance (p > .34). These data support the conclusion that beginning with self-adjustment of high-frequency output resulted in a 2 to 3 dB reduction of self-adjusted low-frequency output.

Figure 4 also shows how the three self-adjustments compared to the group-mean generic starting response and the prescribed NAL-NL2 responses. Relative to the generic starting response, these participants increased overall output by around 10 dB. Between 500 and 2000 Hz they increased spectral slope by 1 dB/octave when adjusting loudness first, and 2 dB/octave when adjusting crispness first. All group-mean self-adjustments were within a few dB of the NAL-NL2 prescribed values between 707 and 4000 Hz but were well below at 250 and 500 Hz. This low-frequency roll-off was already present in the generic starting response.

Individual adjustment logs provided measures of time taken from start to finish of each self-adjustment. An analysis of variance was performed, with replication (1^st and 2^nd adjustment) and method (VLH, VH, HVL) as within-subject factors. On average, the second replication took four seconds less than the first, but the main effect of replication failed to reach the .05 level of significance. The main effect of method, however, was highly significant (F(2,42) = 18.6, p <.0005, ⴄ²_p = 0.47). Collapsed across replication, mean times were 123 sec for 3-parameter crispness-first, 117 sec for 3-parameter loudness-first, and 69 sec for 2-parameter loudness-first. In post-hoc testing, using the Neuman-Keuls test, the difference between the two 3-parameter methods failed to reach the .05 level of significance but the 2-parameter method took significantly less time than did either of the 3-parameter methods (p <.0005).

Individual self-adjustments relative to the NAL-NL2 starting prescription

The output analyses just reported were of absolute real-ear levels in dB SPL. The following analyses examine the difference between each participant’s self-adjustment and his or her NAL-NL2 real-ear target prescription for a 65 dB SPL speech input. For each participant, four measures were obtained: i) the mean of the six self-adjusted high-frequency outputs (the average of 2, 3 and 4 kHz), ii) the range (from highest to lowest) of the six high-frequency adjustments, iii) the mean of the six self-adjusted low-frequency outputs (the average of 500, 750, and 1000 Hz), and iv) the range of the low-frequency adjustments. Age, sex, aid-use, and four-frequency average threshold (0.5, 1, 2 and 4 kHz) were examined as possible predictors of these four individual measures.

High-frequency output relative to NAL-NL2

The data points in Figure 5 show the six self-adjusted high-frequency outputs for each participant. These are arranged in ascending order of the mean of the six adjustments, as shown by the heavy grey line. Vertical lines show the range of the six adjustments. Results for replications 1 and 2 are shown, respectively, to the left and right of these lines. The horizontal axis shows participant number. Participant sex is shown at the top.

Figure 5.

Self-adjusted high-frequency outputs relative to NAL-NL2 prescription.

The means of the six high-frequency self-adjustments for each participant covered a range of 22 dB, from 8 dB above the NAL-NL2 starting prescription to 14 dB below. The group mean was a non-significant 1.1 dB below prescription (SE = 1.2 dB, t(21) = .94, p = .52). Means for 21 of the 22 participants (95%) were within ±10 dB of prescription. Means for 15 (68%) were within ±5 dB.

In analysis of variance with replication and method as within-subject factors at two and three levels, respectively, no main effects or interactions reached the .05 level of significance. There was no evidence that group-mean self-adjustment, relative to the NAL-NL2 prescription, was affected by replication, number of controls, or sequence of controls.

Nevertheless, it is clear from Figure 5 that there are individual differences of high-frequency mean self-adjustment relative to NAL-NL2. Of the four potential predictors mentioned above, sex was the only one showing significance. The average adjustment by the eleven women was 1.3 dB above NAL-NL2. The average adjustment by the eleven men was 3.6 dB below (difference = 4.9 dB, t(20) = 2.24, p = .037). As shown in Figure 5, six of the seven participants with the highest mean were women while six of the seven participants with the lowest mean were men.

It is noteworthy that neither aid-use, nor hearing-loss showed significant correlation with high-frequency self-adjusted output, relative to the NAL-NL2 prescription (Point biserial correlation for aid use = .19, p = .40; product-moment correlation for the four-frequency average threshold =.22, p = .33).

The second variable of interest was the range within the six high-frequency adjustments. This also varied across participants. The average was 7.5 dB (SD = 2.5 dB). The range was from 5 to 12 dB. Of the four predictors, only age showed significant correlation (r(20) = .480, p = .024). The mean range for the eleven listeners younger than 72 was 8.6 dB (SD = 2.1 dB). The range for the eleven listeners aged 72 or above was 5.1 dB (SD = 1.4 dB). The difference was 3.5 dB (t(20) = 3.7, p = .001).

In summary, group-mean high-frequency self-adjustments were not significantly different from NAL-NL2 prescription, but individual adjustments were. Participant differences depended, in part, on sex. The men in this sample tended to select lower outputs than did the women. Consistency, over the six adjustments, also varied, with older participants being more consistent than younger participants.

Low-frequency output re NAL-NL2

Low-frequency self-adjustment data are shown in Figure 6. The means of the six individual self-adjustments (shown by the heavy grey line) cover a range of 31 dB, from 10 dB above the NAL-NL2 prescription to 21 dB below. The group mean was a significant 4.5 dB below prescription (SE = 1.7 dB, t(21) = 2.7, p = .007). Means for 18 of the 22 participants (82%) were within ±10 dB of prescription. Means for 13 (59%) were within ±5 dB.

Figure 6.

Self-adjusted low-frequency outputs relative to NAL-NL2 prescription.

In analysis of variance, with replication and method as repeated measures at 2 and 3 levels, respectively, there was no evidence of a significant effect of replication. There was, however, an effect of method which remained significant after Greenhouse-Geisser correction for non-sphericity (F(1.3,26.9) = 6.3, p = .012, ⴄ²_p = .24). Post-hoc testing isolated this effect to lower output when high frequencies (crispness) were adjusted first. Means for LVH, VH, and HVL were −3.2, −4.2, and −6.0, respectively). These data are in keeping with the absolute dB SPL output results reported earlier.

As with the high-frequency data, consistent individual deviations of average self-adjusted low-frequency output, relative to the NAL-NL2 prescription, are evident. Of the four potential predictors, both hearing loss and age showed significant correlation (r(20) = −.539 and −.446, p = .010 and .037, respectively). Age and hearing loss were not significantly correlated (r(20) = .06, p = .79), indicating that these two variables made independent contributions. Multiple regression analysis showed that the two variables together explained 41% of the variance in low-frequency self-adjusted output (R² = .408, F(2,19) = 8.23, p = .003). Four-frequency average threshold significantly predicted low-frequency adjustment relative to NAL-NL2 (β = −.52), as did age (β = −.42). The age effect is particularly interesting. In the fifteen participants aged below 80, the mean low-frequency adjustment is only a non-significant 1.3 dB below the NAL-NL2 prescription (t(14) = 1.3, p = .21).

The range within each participant’s six low-frequency adjustments varied from a low of 4.0 dB (participant 20) to a high of 26.3 dB (participant 16). This high value, however, was attributable to this participant’s second 3-parameter adjustment starting with crispness and was an outlier when compared with the other participants (d` = 3.1). Participant 16’s range falls to 12 dB when this single observation is excluded. Without this outlying condition, the average range for the 22 participants was 8.2 dB with a standard deviation of 3.3 dB. Among the four predictors (age, sex, aid-use, hearing loss), no correlations reached the .05 level of significance (r <= .15, p > .50).

In summary, these data are in keeping with the earlier finding that group-mean low-frequency adjustments for the 22 participants were significantly below the NAL-NL2 prescription. Low-frequency adjustment relative to NAL-NL2, however, fell with both increasing age and increasing hearing loss. In the participants aged below 80, the mean low-frequency adjustment was not significantly different from the NAL-NL2 prescription for a 65 dB SPL speech input. Ranges of low-frequency self-adjustment were only a little higher than those found for the high-frequencies. Both averaged around ±4 dB relative to the individual mean.

Of note, is the absence of significant correlation between the individual high- and low-frequency self-adjustments relative to NAL-NL2 (r(20) = .089, p = .75). Had participants been adjusting only overall level without attention to spectral slope, a high correlation between the high- and low-frequency self-adjustments would have been expected.

Speech perception

Group-mean Performance vs Intensity (P/I) function

Figure 7 shows group-mean phoneme recognition (±1 SE) as a function of real-ear rms speech level. Real-ear levels are shown with reference to both the generic starting response (circles) or the average of the two final self-adjustments beginning with Volume (triangles). The curves are least-squares fits to a cubed exponential growth function (Boothroyd, 2008).

Figure 7.

Group-mean Performance versus Intensity functions.

Percent scores were arcsine-transformed to increase homogeneity of variance and examined in a repeated-measures analysis. The 0 dB Reference was at two levels (generic-start and self-adjustment). Change was at four levels (+7.3, 0, −7.3, and −14.5 dB relative to the 0 dB reference). After Greenhouse-Geisser correction for non-sphericity, the main effect of reference was significant (F(1,21) = 10.7, p < .004, ⴄ²_p = 0.34). So also, were the main effect of change (F(1.8, 37.6) = 46.3, p < .0005, ⴄ²_p = .69) and the interaction between the two (F(2.0,41.4) = 10.9, p = .0002, ⴄ²_p = .34). Group-mean phoneme recognition varied with level but was better with the self-adjusted output than with the generic starting response.

In Neuman-Keuls post-hoc testing of the interaction between reference and change, group-mean performance with the self-adjusted output was significantly better than with the generic-starting response at 0, −7.3, and −14.5 dB relative to the reference (p <= .003). At 7.3 dB above the reference, however, the difference failed to reach the .05 level of significance. A simple 7.3 increase from the generic-starting response, without change of spectral shape, brought group-mean phoneme recognition to a level that was not significantly different from that obtained either with the self-adjusted response, or with an increase of 7.3 dB relative to self-adjustment (p = .73).

It will be seen from Figure 7 that, for phoneme recognition scores in the region of 80%, the two curves were shifted by around 10 dB. For this group, self-adjustment compensated for the 10 dB high-frequency difference between median high-frequency hearing loss and starting response noted in Figure 1.

Of primary interest, however, is placement of self-adjusted output along the Performance versus Intensity function. Examining only the self-adjusted data, post-hoc testing failed to show a significant difference of group-mean phoneme recognition for an increase or decrease of 7.3 dB relative to self-adjustment. Only when output was reduced by 14.5 dB did a significant drop of phoneme recognition occur (p = .0001). As a group, these participants placed themselves well along the plateau of the Performance versus Intensity function with at least 7 dB to spare in either direction.

Questionnaire Responses

Figure 8 shows the distributions of responses to the items in the questionnaire administered after the second set of self-adjustments.

Figure 8.

Responses to the post-adjustment questionnaire.

Adjustment criteria

Only one participant selected “Effort” as the criterion driving self-adjustment. The other four criteria were selected at roughly equal rates. It is notable that intelligibility was selected by only six of the twenty-two participants. Nevertheless, subsequent comments on the sequence of adjustment were heavily focused on intelligibility.

Instructions

Instructions on the self-adjustment process were provided both verbally and on screen. Only one participant found them to be unclear.

Adjustment number

Although the number of adjustment parameters had no significant effect on group mean output, the questionnaire responses suggest a preference for three. The most common response was “Much prefer three”, selected by 10 participants. Only four selected “No preference”. Of eighteen participants who expressed a preference, twelve preferred having three controls while six preferred two. In a simple χ² test, the difference was significant at the .05 level (χ² = 4.0, p = .05).

Adjustment order

The most frequent response was “No preference”, selected by 10 participants. Of those expressing a preference, six selected Volume-first, and six selected Crispness-first.

Fitting responsibility

The last question provided an opportunity to explore participants’ opinions on the topic of hearing-aid self-fitting – this being the issue that provided the impetus for the research program of which this study is part. Only three participants selected self-fitting without involvement of a professional. Another six selected self-fitting, but with professional involvement. One participant appeared happy to put fitting solely in the hands of a professional and one had no preference. The highest number of selections was for professional fitting with client input – as is the current standard of clinical practice.

Themes in participant comments

Although this was not designed as a qualitative study, participants were given the opportunity to comment on each topic in the questionnaire. These comments provide further insights into their opinions on self-adjustment and self-fitting.

Adjustment criteria

Sixteen participants commented on the question about adjustment criteria. A primary theme (8/16 comments) could be interpreted as lack of criterion independence. For example: “All are important”, “Quality and understanding the talker are the same for me, but all of the options affect listening effort”, “Most important is understanding talker [but] comfort is important because it takes everything else into consideration” and “…comfortable to listen to without straining but loudness is very important”. Some comments revealed individual differences in interpretation of the terminology in the questionnaire. For example: “Quality of sound - didn’t like it too loud or too light” and “Comfort is most important - wouldn’t want to listen to it if it were too loud, but also if it were too soft”.

Adjustment order

Eleven participants commented on the question about order of adjustments. Most comments associated order with adjustment criteria. Justifications for preferring “Crispness” as the initial adjustment included: “Crispness first, because with the voice being clearer to hear, it will help to better understand”, and “…need to adjust crispness to optimize understanding, then adjust amplitude”, and “I can adjust to understand but then maybe need more loudness”. Justifications for preferring “Loudness” as the initial adjustment included: “Loudness is my primary problem - hearing soft sounds. I like to take care of that first and then adjust the others” and “Loudness (much preferred) is the one you usually do very first, even on [one’s] own hearing aids”. One participant was quite adamant “Crispness should never be the first choice”.

Adjustment number

Among ten comments on number of parameter adjustments, a preference for two appeared to be related to the theme of simplicity. For example: “Easier with two” and “Three things to adjust are confusing”. A preference for three adjustments appeared to be related to the theme of control. For example: “…liked having all three because it gets …closer to the best ability to hear” and “The more the better.”

Responsibility for aid fitting

Fourteen participants commented on the issue of hearing aid fitting. A primary theme (8/14) related to knowledge and experience. For example: “The professional should know more than I do”, “I certainly don’t have the experience to do it alone” and “Hearing professional could point out recommendations that I would/might miss”. Less common (3/14) were comments relating to the theme of control. For example, “It’s gonna be my ear that’s hearing, so I would like to give major input” and “I like being responsible for myself”. The theme of cost appeared only twice: “it also depends on how much the professional would charge” and “If I can get help for a couple of hundred dollars with me being the ‘pretend audiologist’ and a back-up ‘real audiologist’ to help me if I need it, then that would be great”.

Discussion

The primary purpose was to determine the effects of number and sequence of adjustment parameters. These data do not support the conclusion that changing from a 3-parameter method (volume, treble, bass) to a 2-parameter method (volume, spectral tilt) influenced individual or group mean self-adjustments. Nevertheless, twelve participants expressed preference for three adjustment parameters while only six preferred two. The fact that the two-control method reduced mean adjustment time from around 2 minutes to around 1 minute does imply a benefit but there was no evidence that this factor influenced the participants.

Beginning the three-parameter method with adjustment of high-frequency output rather than overall amplitude resulted in reduced low-frequency self-adjusted output. The change was of the order of one 3 dB adjustment step. The absence of an effect above 1 kHz suggests that self-adjustments were driven by high-frequency output, which further suggests intelligibility as a major factor. The relative contributions of intelligibility and other adjustment criteria, however, are not clear from output spectra alone.

The principal conclusion to be drawn from these data is that the explore-and-select approach to self-adjustment is relatively immune to changes of the number and sequence of the adjustment parameters – at least, during initial exposure. Note, however, that these participants were first presented with one adjustment parameter at a time. This study provides no information on self-adjustment when all controls are made available at once.

There was no evidence that group-mean self-adjusted output was affected by replication. This result differs from that of the earlier study in which aid users showed a replication effect. (Mackersie at al., 2019). Note, however, that there was a difference in instructions. In an attempt to ensure adequate exploration before selection, participants in the present study were asked to increase each parameter until it was “higher than you want”, and to decrease until it was “lower than you want”, before adjusting to preference. In the earlier study, they were asked only to “adjust … to your liking”. Note, also, that the present study provided three opportunities for self-adjustment within each replication. In other words, the first of the two “replications” was not a single trial but already included three self-adjustments, using different protocols.

The third research question dealt with the relationship between self-adjustment and the NAL-NL2 threshold-based starting prescription. Group-mean self-adjustments, and many individual adjustments, were close to this prescription. Recall that the adjustments reported here did not start with these individual prescriptions but with a generic prescription based on a typical mild-to-moderate hearing loss. Nevertheless group-mean outputs, averaged over six self-adjustments, were only 1.3 dB below the average NAL-NL2 prescription in the high frequencies and 3.6 below in the lows. Note that this last figure fell to 1.3 dB for the fifteen participants younger than 80. These data speak well for the individual prescription as a means of placing listeners as close as possible to what will be their preferred response after fine tuning. It also suggests, however, that a generic starting response, such as the one used here, could be effective for over-the-counter, self-fitting hearing aids.

Although group-mean self-adjustments were close to mean prescription, individual differences were considerable. As shown in Figures 5 and 6, the full range of preferred high-frequency outputs (averaged over six self-adjustments and relative to NAL-NL2 prescription) ranged from +8 to −14 dB in the high frequencies and +10 to −21 dB in the lows. Nelson et al., (2018) reported similar ranges for self-adjustment in quiet (+8 to −20 dB in the high frequencies and +18, to −21 dB in the lows). The two studies used a similar explore-and-select procedure but with very different hardware and software. As indicated in the introduction, the software used by Nelson et al., employed a sophisticated algorithm in which changes of overall volume were accompanied by changes of frequency response and compression characteristics.

Despite the range of differences between self-adjusted and prescribed output, it is noteworthy that 68% of the participants in the present study self-adjusted to within ±5 dB of prescription in the high frequencies, which are of most importance to intelligibility. The data are less favorable in the low frequencies, but even there, just over 59% met this criterion. In the Nelson et al., data, the corresponding values (from their Figure 6) are similar, at 73% and 47%, even though “high” and “low” were defined somewhat differently in the two studies.

The finding that the women’s real-ear self-adjustment relative to the NAL-NL2 prescription was some 5 dB higher than that of the men was surprising and differs from that of Perry et al., (2019). The mean NAL-NL2 real-ear output prescriptions generated by Verifit did not differ significantly for the two groups. However, these prescriptions used a generic real-ear to coupler difference which does not take account of sex-related ear-acoustics. Had the smaller volumes of the female ear been accounted for, the women’s target outputs in the high frequencies could have been higher by 1 to 2 dB (Dillon, 2012, p. 97) and the deviations from target would have been a little lower. The effect observed here, however, could simply reflect the characteristics of these participants. The samples of men and women were by no means balanced.

The fourth research question dealt with repeatability of self-adjustment. Within-participant repeatability over six adjustments was remarkably good. The group-mean range of adjustments was around ±4 dB in both the high and the low frequencies. Some participants were as low as ±1 dB and none was above ±6 dB. The source of variability within each participant’s six self-adjustments, however, is not clear from these data. It could be that participants could not detect differences among the six adjustments. To the extent, however, that each adjustment ended with acceptance, the variability could also reflect a range, or zone, of acceptable listening, as suggested by Dreschler et al. (2008). If so, there is evidence from the present data that older listeners have a lower range of acceptability in the high frequencies. The source and significance of variability in repeated adjustment is an issue in need of further exploration. What is clear from these data, however, is that individual differences in group-mean self-adjustment are not simply the consequences of within-participant variability. There are genuine between-participant differences of preference - hence the need for fine-tuning.

The fifth research question dealt with speech perception before and after self-adjustment. The group-mean Performance/Intensity function of Figure 7 shows that self-adjustment reached a level providing close to maximal phoneme recognition. This finding adds support to the conclusion that intelligibility played a major role. It does not appear, however, that participants sought the lowest possible level consistent with adequate intelligibility. Group-mean performance remained at or above 90 % when input was reduced from self-adjustment by 7.3 dB. Of the individuals who fell below a 90% criterion at −7.3 dB, only three scored less than 80%. A phoneme recognition score of 80% is compatible with over 96% recognition of words in “relatively easy” sentence material without benefit of narrative context (Boothroyd, 2008). These intelligibility findings are encouraging, but they do not guarantee that output change relative to self-selection will have no effect on other factors such as listening effort, fatigue, or long-term acceptability.

The phoneme recognition data and the high-frequency self-adjustment data support the notion that use of the NAL-NL2 prescription as a starting response leads to the least amount of fine-tuning for the highest number of listeners. It would not, however, eliminate the need for fine-tuning. This need exists for audiologist fitting, regardless of whether self-adjustment is included. The need is especially acute for self-fitting of over-the-counter hearing aids. Even if algorithms for self-audiometry are provided in the aid, to be used as a basis for prescription, their implementation adds time and potential inaccuracies (Convery et al., 2017).

It is also important to distinguish between fine tuning during initial fitting and post-fitting self-adjustment. The latter enables listeners to respond to changes of source, interference, and listening demands in the real world. The industry’s recognition of this distinction has led to numerous smart-phone applications for post-fitting self-adjustment.

The questionnaire responses and subsequent comments serve to emphasize individual differences – not only of amplification needs and self-adjustment capabilities, but also of user-interface preferences and locus of control. A preference for high-frequency adjustment before volume adjustment could indicate the importance of intelligibility. It could also, however, reflect the influence of steeper loudness versus amplitude functions in the higher frequencies. Intelligibility was not reported as the most important criterion. Indeed, comments on the question about subjective adjustment criteria revealed participant awareness of non-orthogonality among the dimensions offered. More importantly they serve to illustrate a major issue affecting hearing-aid fitting. Namely, while loss of intelligibility is often the primary driver in the search for assistance, adjustment of output and spectrum becomes a compromise among such things as intelligibility, comfort, listening effort, sound quality, noise-tolerance, and convenience.

Responses to the last question on the relative balance of self-fitting and professional fitting are particularly interesting. The high number of selections of “professional with input from me” is only partly explained by the fact that this reflects the experience of current standard of practice by the ten participants who already wore hearing aids. The other twelve participants were current non-users. At first sight the small number of participants selecting “just me” does not augur well for self-fitting of direct-to-consumer hearing aids. This was not, however, a random sample of potential customers for such aids. As just indicated, ten of these participants had already acquired hearing aids from professionals, and the fact that the other twelve volunteered to participate in a research study implies trust of professionals.

It must be emphasized that this study was not about complete hearing-aid self-fitting, which includes such components as assembly, insertion, and operation (Convery et al, 2019). The study dealt only with self-adjustment of output and frequency response. The goal of the research of which this study formed part was to assess the efficacy of, and candidacy for, self-adjustment of amplification. The data on proximity of self-adjustment to the average data, implicit in the NAL-NL2 starting prescription, adds to a growing body of evidence on efficacy, as does the high level of repeatability within individuals. But the present study provides no information on self-adjustment of other factors such as compression and maximum output, and the present findings provide little information on candidacy. The low-frequency deviations from the NAL-NL2 starting prescription suggest a greater likelihood of difficulty for listeners over 80 but age alone would not serve as an exclusionary criterion.

This was a laboratory study using preprocessed stimuli and computer presentation via an insert earphone. As a hearing-aid simulation, the effective input was constant, noise was minimal, and no compression was invoked by listener adjustment of output at any frequency. It was assumed that participants would treat the adjustment task as if they were listening to average conversational speech after amplification. The NAL-NL2 prescription of ear-level output for a 65 dB SPL speech input was, therefore, chosen as a reference against which to compare the results of self-adjustment. Proximity of group-mean outputs to this prescription supports that choice and speaks well for the NAL-NL2 prescription. Future studies will need to include real-time processing of a variety of sound-field inputs, the inclusion of amplitude compression, and validation for more than one input level.

Because this study involved computer output of pre-processed sound stimuli, it avoided the issues of acoustic feedback and sound-field input past or through receiver domes that would have occurred with actual hearing aids. The issue of low-frequency leakage from the ear canal, past or through receiver domes, was still present and could have contributed to the low-frequency differences between self-adjusted and NAL-NL2 prescribed output, especially in the older participants.

As mentioned earlier, the proximity of group self-adjustment to the NAL-NL2 high-frequency prescriptions puts into question the need for self-adjustment to begin from such prescriptions. Such a conclusion is counter to the results of studies in which final self-adjustment is dependent on the starting response (Dreschler et al., 2008; Keidser et al., 2008). In the present study, however, the generic starting response was close to the individual prescription for many participants. A direct comparison of adjustments starting with this generic response and individually prescribed responses was not done here. However, the distribution of outputs relative to the NAL-NL2 prescription was similar to that found by Nelson et al., (2018) using NAL-NL2 as the starting response and the Ear Machine software. It may well be that one or more generic starting responses that match the needs of typical candidates for over-the-counter hearing-aids will be adequate, assuming that users are provided with a means to explore, and select from, a range of self-adjustments.

No attempt was made in this study to bias participants towards a specific criterion for self-adjustment. The criterion of “best for you” was chosen intentionally. As pointed out by these participants and by many writers, (e.g., Preminger & Van Tassell, 1995; Jensen et al., 2019), selection involves a compromise among multiple, often conflicting, criteria which can differ between individuals and even within an individual as he or she responds to changing conditions and communication demands. In fact, inter- and intra-participant difference in making this compromise may well be responsible for some of the differences in self-adjusted output reported here.

Conclusions

1. Outcomes of the Goldilocks explore-and-select approach to self-adjustment were relatively insensitive to changes of number and sequence of adjustments.

2. The change from three to two controls did, however, reduce average adjustment time from around 2 minutes to around 1 minute.

3. Many individual self-adjustments were close to the NAL-NL2 starting prescription, but the full range was from +8 to −14 dB in the high-frequencies and +10 to −21 dB in the lows.

4. Over six self-adjustments, the average participant stayed within ± 4 dB of his or her average in both the high and low frequencies.

5. There was no evidence that hearing-aid users and non-users differed in their self-adjustment relative to NAL-NL2 prescription.

6. Group-mean self-adjustments placed listeners along the plateau of a Performance vs. Intensity function with around 7 dB to spare in either direction.

7. Participants differed in terms of preference for the number and sequence of adjustment parameters.

8. These differences reflected the inter-dependence of intelligibility, loudness, comfort, and sound quality as well as conflicting desires for greater control versus greater simplicity.

Overall, the findings continue to support the conclusion that a high proportion of adults with mild-to moderately severe hearing loss are capable of using an explore-and-select procedure to repeatedly arrive at a hearing-aid output response close to the NAL-NL2 starting prescription, but self-adjustments beyond amplitude and spectral slope remain to be explored.