Dynamically Masked Audiograms With Machine Learning Audiometry

Abstract

Objectives: When one ear of an individual can hear significantly better than the other ear, evaluating the worse ear with loud probe tones may require delivering masking noise to the better ear in order to prevent the probe tones from inadvertently being heard by the better ear. Current masking protocols are confusing, laborious and time consuming. Adding a standardized masking protocol to an active machine learning audiogram procedure could potentially alleviate all of these drawbacks by dynamically adapting the masking as needed for each individual. The goal of this study is to determine the accuracy and efficiency of automated machine learning masking for obtaining true hearing thresholds.Design: Dynamically masked automated audiograms were collected for 29 participants between the ages of 21 and 83 (mean 43, SD 20) with a wide range of hearing abilities. Normal hearing listeners were given unmasked and masked machine learning audiogram tests. Listeners with hearing loss were given a standard audiogram test by an audiologist, with masking stimuli added as clinically determined, followed by a masked machine learning audiogram test. The hearing thresholds estimated for each pair of techniques were compared at standard audiogram frequencies (i.e., 0.25, 0.5, 1, 2, 4, 8 kHz).Results: Masked and unmasked machine learning audiogram threshold estimates matched each other well in normal hearing listeners, with a mean absolute difference between threshold estimates of 3.4 dB. Masked machine learning audiogram thresholds also matched well the thresholds determined by a conventional masking procedure, with a mean absolute difference between threshold estimates for listeners with low asymmetry and high asymmetry between the ears, respectively, of 4.9 dB and 2.6 dB. Notably, out of 6200 masked machine learning audiogram

Automated and machine learning approaches in diagnostic hearing assessment: a scoping review

“…This behavioral information can be used to develop algorithms that detect loss of engagement or lack of attention and create a warning system to draw the listener back to the task. Schlittenlacher et al, 2018 20 Bayesian active-learning methods provide an accurate estimate of hearing thresholds in a continuous range of frequencies Schmidt et al, 2014 43 A user-operated two-alternative forced-choice in combination with the method of maximum likelihood does not require specific operating skills and the repeatability is acceptable and similar to conventional audiometry Song et al, 2015 18,[62][63][64]…”

“…Machine learning methods converged to a 5 dB precision for unmasked air conduction audiograms within 18 to 28 trials for one side 20 or 19 to 44 trials bilaterally. 63 The gold standard method required 93 to 114 trials on average, depending on the hearing loss configuration. 63 Heisey et al 63 also expressed the efficiency gains by comparing the total overall test time of 2•1 to 4•8 minutes for bilateral air conduction audiometry using the machine learning based procedure versus 6•9 to 9•9 minutes using the gold standard procedure.…”

“…63 The gold standard method required 93 to 114 trials on average, depending on the hearing loss configuration. 63 Heisey et al 63 also expressed the efficiency gains by comparing the total overall test time of 2•1 to 4•8 minutes for bilateral air conduction audiometry using the machine learning based procedure versus 6•9 to 9•9 minutes using the gold standard procedure. The two maximum likelihood approaches reported test time per ear 27 or per frequency.…”

“…In fully adaptive procedures, the level of precision and confidence needed to conclude the assessment can be set to any level by choosing the proper termination criteria, resulting in different trade-offs. Schmidt et al, 43 for instance, aimed for high accuracy and reliability, whereas Heisey et al 63 aimed for high efficiency. Overall, several different measures were used to demonstrate accuracy across the identified reports.…”

On a Vector towards a Novel Hearing Aid Feature: What Can We Learn from Modern Family, Voice Classification and Deep Learning Algorithms

“…The features include, but are not limited to, directional microphones [2][3][4][5][6][7][8][9][10] noise reduction algorithms [11], dynamic range compression [12] and proper fitting and verification [13][14][15][16]. What is notable is that each of these factors was conceptualized, tested, validated and incorporated into current hearing aid devices via digital signal processing (DSP) techniques, advanced hearing aid technology, and more recently, machine learning approaches [17][18][19][20]. In each case, the incorporated features are associated with increased identification, assessment and/or hearing performance across a variety of listening environments [17,[19][20][21][22].…”