Existing guidelines for hearing aid (HA) fitting predominantly focus on frequency-specific audibility compensation for hearing loss.1,2 Current HA fitting practices for advanced features such as compression speeds, noise reduction, and frequency lowering rely on manufacturer defaults, individual expertise, or subjective patient feedback.3 These HA features may interact with each other4 as well as with the listener's environment, resulting in cumulative changes to the fidelity of the signal5,6 (e.g., speech envelope distortions) beyond the expected improvements in audibility. Without established guidelines for selecting and fine-tuning these features, there may be a risk of suboptimal HA outcomes, especially for hard-of-hearing listeners who are disadvantaged (e.g., older adults, poorer cognitive abilities) by more advanced settings of these features (e.g., faster compression speeds or greater noise reduction), particularly in challenging situations such as high background noise.7,8 Other than the Speech Intelligibility Index, which is a measure of audibility available on some clinical HA verification equipment, no objective metric is readily accessible to audiologists to measure overall HA signal fidelity. To facilitate evidence-based HA fitting, beyond achieving adequate gain, audiologists need tools to capture the acoustic effects of combined HA feature adjustments in the context of the HA user's listening environment. Moreover, for a truly individualized fitting, these tools should account for individual ear acoustics and vent effects that, in turn, influence the effectiveness of HA processing.9
Recent work has shown that it is possible to quantify signal fidelity from clinically fit HAs5,6 using cepstral correlation,10,11 which measures changes to the speech envelope of a processed signal and is a reliable predictor of speech intelligibility and quality across several forms of HA processing.7,8,12 Using this metric, Rallapalli, et al.,6 measured signal fidelity across 96 clinically prescribed adult HA fittings by recording HA output in response to different levels of speech and background noise on an acoustic manikin. The study also demonstrated the advantages of an acoustic manikin over a test box to capture the open-ear effects and vent acoustics associated with commonly prescribed HAs, including receiver-in-the-canal type.
Although an acoustic manikin is representative of the average human ear,13 it may under- or overestimate the ear canal and pinna acoustics for some individuals.14 Moreover, an acoustic manikin may not be a readily accessible and cost-effective option in the clinic. An ideal clinical tool is one that can measure the acoustic effects of HA feature adjustments directly in a patient's ears. It is also important that such a tool is convenient and time-efficient to be successfully implemented in the clinic. This study takes a step in this direction and proposes the use of a commonly available clinical equipment, Audioscan Verifit 2 (VF-2), to overcome some of these issues. This study aims to determine the feasibility of measuring HA signal fidelity at user settings of recordings obtained from individual ears, with the VF-2.
Participants & Hearing Aids: Participants were four adults (69-80 years old; three men) with bilateral sensorineural hearing loss. Figure 1 (A) shows air conduction thresholds for the test ears. None of the participants reported having otologic problems or occluding cerumen at the time of testing.
The participants’ right HAs were used in the study. Three participants wore receiver-in-the-canal HAs, two with open domes, one with a closed dome, and one participant wore a completely-in-the-canal HA. Before the recording, fresh batteries were placed in the HAs, and a listening check was performed. Real-ear aid responses (REAR) were measured at 55, 65, and 75 dB SPL input levels against NAL-NL2 prescriptive targets15 using a male talker signal (“carrot passage”) from the VF-2. Figure 1 (B-E) shows the match-to-targets. No adjustments were made to the participant's HAs because the study focused on determining the feasibility of the recording methodology at user settings. The match-to-NAL-NL2 targets in this study were consistent with other reports of real-world clinical fittings.16 All participants completed an informed consent process approved by the Institutional Review Board at Northwestern University.
Stimuli & Listening Conditions: Similar to Rallapalli, et al.,6 stimuli were two sentences,17 spoken by one male and one female talker. To capture the HA response at realistic background noise levels for hard-of-hearing listeners,18 the researchers mixed sentences with six-talker babble19 at three signal-to-noise ratios (SNRs; 5 dB, 10 dB, and quiet). To capture the effects of soft, average, and loud speech, noisy and quiet stimuli were presented at three input levels (55 dB, 65 dB, 75 dB SPL), measured at two feet from the VF-2 loudspeaker.
Recording Setup: Recordings were completed in a double-walled sound-treated booth (Fig. 2), representative of a typical clinical environment. The calibrated probe microphone was placed in the right ear at depths (from inter-tragal notch) of 28 mm and 30 mm for female and male participants, respectively. Stimuli were presented using the USB connection of the VF-2, with each sound file consisting of a single repetition of the stimulus at each SNR. The stimuli and the corresponding input level were selected using the dropdown menu in the on-ear Speechmap mode. The output from the probe microphones was routed (via VF-2’s headphone monitoring jack) through a digital-to-analog converter (M-Audio 8) and saved in an external PC. The researcher was seated outside the booth and used a custom Matlab program to capture 90-second recordings (about six repetitions) in each listening condition (three SNRs X three input levels). The repetitions were later averaged to obtain the final recording that was subsequently used to calculate the signal fidelity in each condition, as described below.
Procedure: Participants were instructed according to standard real-ear measurement procedures. They were also instructed to minimize movements for the duration of the recording. First, unaided recordings were obtained in quiet at each input level to serve as an individual participant's reference signal for the metric. This would ideally allow us to account for individual ear acoustics. Next, aided recordings were obtained at user settings for all listening conditions. The researcher monitored the recording process and inspected each recorded waveform. If a given recording was deemed noisy (e.g., sneeze or sudden movement), it was re-recorded (and the participant reinstructed as needed). The total duration of the recording was approximately 18 minutes per participant: 4.5 minutes (90 s X three levels; unaided) + 13.5 minutes (90 s X three levels X three SNRs; aided). The entire session, including rest breaks, lasted about one hour.
Signal Fidelity Metric: Cepstral correlation from the Hearing Aid Speech Quality Index (HASQI10) was used to calculate signal fidelity by comparing short-term envelope modulations of the reference signal with the signal processed through the HA. The reference signal was the recording for speech in quiet obtained at each input level. Both the reference and HA processed signals were passed through a peripheral-impaired auditory model that estimated loss of audibility and broadened auditory filters. NAL-R linear gain was applied to the reference signal. As such, the metric captures cumulative signal fidelity as a result of HA processing and the listening condition (background noise and input level), after accounting for individual hearing impairment and HA gain. Any condition that distorts the speech envelope will result in a mismatch between the processed and reference signals, thereby decreasing the cepstral correlation/signal fidelity. Cepstral correlation values range from 0 (no match between envelope modulations of processed and reference signals; least signal fidelity) to 1 (perfect match between the signals; maximum signal fidelity). Full details regarding the metric are available in Kates and Arehart.10,11
Comparison of Recording Methods: Manikin v. Individual Ears: First, a subset of four participants with matched audiometric thresholds (within +/- 10 dB between 0.25-6 kHz) was selected from the Rallapalli, et al.,6 study to determine whether the metric would be affected by recording method (individual ears+VF-2 vs. manikin). A one-way ANOVA showed no significant effect of the recording methodology on the measured signal fidelity (F[1,70]=1.30, p=0.258). This result suggested that signal fidelity could be measured reliably on HA recordings obtained directly from the patient's ears using the VF-2.
Signal Fidelity Across Listening Conditions: The distribution of signal fidelity across SNRs and input levels is shown in Figure 3. Data from the study by Rallapalli, et al.,6 (manikin) are shown for comparison. As expected, signal fidelity decreased as SNR decreased from quiet to 5 dB and as input level increased from 55 dB to 75 dB SPL. The following analysis pertains to signal fidelity measured on individual ears in the present study. A statistical model (linear mixed-effects) showed significant main effects of SNR (F[2,24]=10.53, p<0.001) and input level (F[2,24]=24.13, p<0.001) on signal fidelity. Because the interaction between SNR and input level was also significant (F[4,24]=3.31, p=0.027), paired t-tests were conducted to determine the effect of input level at each SNR. After correcting for multiple comparisons,20 results showed that increasing the input level resulted in decreasing signal fidelity only when noise was present. For example, at 10 dB SNR, increasing the input level from soft (55 dB) to loud (75 dB SPL) resulted in a decrease in signal fidelity by 0.126 (p<0.001). The effect was greater at 5 dB SNR, such that the same increase in input level resulted in a decrease in signal fidelity by 0.217 (p<0.001). This worsening of signal fidelity is likely a result of the HA compressor engaging to a greater extent at high speech and noise levels, causing further distortion of the processed speech signal (envelope) compared with engagement at softer levels. On the other hand, no significant effect of input level was found on signal fidelity in quiet (p>0.05). The overall pattern of results indicates that the methods used in this study captured the behavior of HA processing in the context of clinically relevant speech and noise conditions.
This study established the feasibility of measuring HA signal fidelity under clinical conditions. Consistent with previous work using an acoustic manikin,6 the methods presented in this study captured the pattern of HA response (at user settings) across a realistic range of speech and background noise levels. Although the present study only included a small number of participants and different HAs and HA parameters, the signal fidelity from individual ears in this study was generally comparable to the signal fidelity range noted by Rallapalli, et al.,6 using the acoustic manikin (Fig. 3). Thus, the new recording method using easily accessible clinical equipment, such as the VF-2, enables reliable measurement of combined acoustic effects of HA processing and listening conditions while allowing for individual ear acoustics and vent effects. While this study focused on measuring HA signal fidelity using the cepstral correlation metric, the recording methods introduced here can also be used in other types of analyses.
Several steps are being taken to make the recording methods and signal fidelity metric more clinic-friendly. Work is also underway to reduce potential sources of noise during recording to ensure sufficient fidelity of the recorded signal so it can be used not only in a sound-treated booth but also in a relatively noisier fitting room.21 The ultimate goal is to provide audiologists with an objective tool for guiding HA adjustments.
ACKNOWLEDGMENTS: The author thanks Gregory Ellis for assistance with the recording setup and Pamela Souza for her critiques on this research. The author also thanks Kathryn Arehart, Melinda Anderson, and James Kates for helpful discussions about this project. This work was supported by the National Institutes of Health Grant R01 DC012289 (to Pamela Souza). The author has no conflicts of interest to disclose.
Thoughts on something you read here? Write to us at HJ@wolterskluwer.com.
1. ASHA. Preferred practice patterns for the profession of audiology. American Speech-Language-Hearing Association. 2009. Retrieved July 22, 2019, from www.asha.org/policy
2. Valente M, Abrams H, Benson D, Chisolm T, Citron D, Hampton D, Sweetow R. Guidelines for the audiologic management of adult hearing impairment. Audiology Today
3. Anderson MC, Arehart KH, Souza PE. Survey of current practice in the fitting and fine-tuning of common signal-processing features in hearing aids for adults. Journal of the American Academy of Audiology
. 2018 Feb 1;29(2):118-24.
4. Brons I, Houben R, Dreschler WA. Acoustical and perceptual comparison of noise reduction and compression in hearing aids. Journal of Speech, Language, and Hearing Research
. 2015 Aug;58(4):1363-76.
5. Kates JM, Arehart KH, Anderson MC, Muralimanohar RK, Harvey Jr LO. Using Objective Metrics to Measure Hearing-Aid Performance. Ear and Hearing
. 2018 Nov;39(6):1165.
6. Rallapalli V, Anderson M, Kates J, Balmert L, Sirow L, Arehart K, Souza P. Quantifying the Range of Signal Modification in Clinically Fit Hearing Aids. Ear and Hearing
. 2020 Mar 1;41(2):433-41.
7. Souza PE, Arehart KH, Shen J, Anderson M, Kates JM. Working memory and intelligibility of hearing-aid processed speech. Frontiers in Psychology
. 2015 May 7;6:526.
8. Souza P, Arehart K, Schoof T, Anderson M, Strori D, Balmert L. Understanding Variability in Individual Response to Hearing Aid Signal Processing in Wearable Hearing Aids. Ear and Hearing
. 2019 Nov 1;40(6):1280-92.
9. Winkler A, Latzel M, Holube I. Open versus closed hearing-aid fittings: a literature review of both fitting approaches. Trends in Hearing
. 2016 Feb 12;20:2331216516631741.
10. Kates JM, Arehart KH. The hearing-aid speech quality index (HASQI) version 2. Journal of the Audio Engineering Society
. 2014 Mar 20;62(3):99-117.
11. Kates JM, Arehart KH. The hearing-aid speech perception index (HASPI). Speech Communication
. 2014 Nov 1;65:75-93.
12. Arehart K, Souza P, Kates J, Lunner T, Pedersen MS. Relationship between signal fidelity, hearing loss and working memory for digital noise suppression. Ear and Hearing
. 2015 Sep;36(5):505.
13. Burkhard MD, Sachs RM. Anthropometric manikin for acoustic research. The Journal of the Acoustical Society of America
. 1975 Jul;58(1):214-22.
14. Fikret-Pasa S, Revit LJ. Individualized correction factors in the preselection of hearing aids. Journal of Speech, Language, and Hearing Research
. 1992 Apr;35(2):384-400.
15. Keidser G, Dillon H, Carter L, O'Brien A. NAL-NL2 empirical adjustments. Trends in Amplification
. 2012 Dec;16(4):211-23.
16. Sanders J, Stoody T, Weber J, Mueller HG. Manufacturers' NAL-NL2 fittings fail real-ear verification. The Hearing Review
. 2015 Mar;21(3):24-30.
17. Nilsson M, Soli SD, Sullivan JA. Development of the Hearing in Noise Test for the measurement of speech reception thresholds in quiet and in noise. The Journal of the Acoustical Society of America
. 1994 Feb;95(2):1085-99.
18. Smeds K, Wolters F, Rung M. Estimation of signal-to-noise ratios in realistic sound scenarios. Journal of the American Academy of Audiology
. 2015 Feb 1;26(2):183-96.
19. Cox RM, Alexander GC, Gilmore C. Development of the Connected Speech Test (CST). Ear and Hearing
. 1987 Oct;8(5 Suppl):119S-26S.
20. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal statistical society: series B (Methodological)
. 1995 Jan;57(1):289-300.
21. Clark JG, Brady M, Earl BR, Scheifele PM, Snyder L, Clark SD. Use of noise cancellation earphones in out-of-booth audiometric evaluations. International Journal of Audiology
. 2017 Dec 2;56(12):989-96.