| [Download PDF]
|Year : 2003 | Volume
| Issue : 21 | Page : 17--37
An approach to the development of hearing standards for hearing-critical jobs
C Laroche1, S Soli2, C Giguere1, J Lagace1, V Vaillancourt1, M Fortin1,
1 University of Ottawa, Faculty of Health Sciences, School of Rehabilitation Sciences, Audiology and Speech-Language Pathology Program, Ontario, Canada
2 House Ear Institute, Los Angeles, U.S.A.
University of Ottawa, Faculty of Health Sciences, School of Rehabilitation Sciences, Audiology and Speech-Language Pathology Program, 451 Smyth road, Ottawa, Ontario, Canada, K1H 8M5
Many jobs at the Department of Fisheries and Oceans Canada (DFO) have several features in common: they are often performed in noisy environments and involve a number of auditory skills and abilities, such as speech communication, sound localization, and sound detection. If an individual lacks these skills and abilities, it may constitute a safety risk for this individual, as well as for fellow workers and the general public. A number of scientific models have been developed to predict performance on these auditory skills based on diagnostic measures of hearing such as pure-tone audiograms. While these models have significant scientific and research value, they are unable to provide accurate predictions of real life performance on auditory skills necessary to perform hearing-critical jobs. An alternative and more accurate approach has been developed in this research project. A direct measure of functional speech perception in noise (Hearing in Noise Test: HINT) has been identified and validated for use in screening applicants for hearing-critical jobs in DFO. This screening tool has adequate and well-defined psychometric properties (e.g. reliability, sensitivity, and validity) so that screening test results can be used to predict an individual's ability to perform critical auditory skills in noisy environments, with a known degree of prediction error. Important issues must be considered when setting screening criteria. First, the concept of hearing-critical tasks must be reviewed, since these tasks are often performed in high noise levels where normally-hearing people cannot hear adequately. Second, noise-induced hearing loss is frequent in these noisy environments, and workers who acquire a hearing loss might not continue to meet the minimal auditory screening criteria throughout their career. Other senses (e.g., vision, touch) also play an important role in these environments. Third, adaptation strategies have to be considered when recruits or incumbents fail the screening test.
|How to cite this article:|
Laroche C, Soli S, Giguere C, Lagace J, Vaillancourt V, Fortin M. An approach to the development of hearing standards for hearing-critical jobs.Noise Health 2003;6:17-37
|How to cite this URL:|
Laroche C, Soli S, Giguere C, Lagace J, Vaillancourt V, Fortin M. An approach to the development of hearing standards for hearing-critical jobs. Noise Health [serial online] 2003 [cited 2020 Aug 8 ];6:17-37
Available from: http://www.noiseandhealth.org/text.asp?2003/6/21/17/31684
In the last ten years, human rights commissions and health professionals worldwide have put pressure on employers to use valid and reliable tools to evaluate hearing abilities required by employees in hearing-critical jobs, such as police officers, firefighters, and coast guard employees (Hetu, 1993; Laroche, 1994; MacLean, 1995; Forshaw and Hamilton, 1997; Forshaw et al. 1999; Ritmiller et al., 1999; Soli and Vermiglio, 1999; Bherer et al., 2002). For example, the Canadian Human Rights Commission (CHRC) has asked the Department of Fisheries and Oceans Canada (DFO) to review their hearing criteria because of a discrimination complaint in 1995 by a recruit who was refused a job on the basis of an asymmetrical hearing loss. The DFO criteria for hearing were based on audiometric thresholds. The CHRC Tribunal determined that this type of evaluation was not valid because it only measured hearing sensitivity and not other hearing abilities, such as speech perception in noise or localization of sound sources, required by the job. The DFO was not able to demonstrate that their criteria met the bona fide occupational requirement (BFOR). The Canadian Coast Guard (CCG) and Conservation and Protection (C&P), Sections of Fisheries and Oceans Canada, thus had to review their medical standards to better define critical components of medical fitness essential to safe and effective marine operations.
Most CCG and C&P jobs are hearing-critical (HC), and share several features. They are often performed in noisy environments, and involve a number of common functional hearing abilities such as speech communication, sound detection and localization. Individuals who lack these abilities may constitute a safety risk to themselves, to fellow workers, and to the general public. The fact that these tasks are often performed in noisy environments means that people in these jobs can acquire noise-induced hearing loss over their career. Noise-induced hearing loss is the most prevalent irreversible worldwide occupational hazard (WHO, 2000). In the United States, it is estimated that 1 million workers suffer from work-related hearing loss (Weeks et al., 1991). When determining minimal hearing criteria to perform specific auditory tasks, no one can neglect these aspects. At hiring, a candidate can have normal auditory abilities, but the workplace noise can jeopardize these abilities throughout the years. The tasks do not necessarily change throughout these years; they are still hearing-critical, but workers have lost part of their hearing capacities.
Currently, job requirements related to auditory abilities are almost always based on medicolegal definitions of hearing, such as average hearing thresholds at specific frequencies. These criteria were originally designed for compensating people with noise-induced hearing loss (Hetu, 1993), not to predict functional hearing abilities. A number of quantitative models have been developed to predict functional hearing ability from diagnostic measures of hearing such as the pure-tone audiogram (Pavlovic et al., 1986; Forshaw & Hamilton, 1997; ANSI S3.5-1997; Forshaw et al., 1999). While these prediction models may operate very well on the average and have research value, they lack sufficient accuracy to predict real-life performances of specific individuals from which decisions and actions regarding HC jobs can be based (Coles & Sinclair, 1988; Hetu, 1993; Laroche, 1994; Begines, 1995; MacLean, 1995; Forshaw & Hamilton, 1997; Bherer et al., 2002). For example, the models would predict identical speech intelligibility for individuals with identical pure-tone thresholds, whereas a substantial variability of 5-10 dB in speech recognition thresholds is found in practice (Smoorenburg, 1992; Soli and Vermiglio, 1999; Killion and Niquette, 2000).
A potentially more accurate alternative approach has been used in present research. Diagnostic measures of hearing have been replaced with simple computerized screening measures of functional hearing in reference noises. These screening measures can be administered quickly over headphones. Sound field measures are also available in cases where hearing aids or other devices are worn. Through statistical modeling, performance on the screening measures exhibits a direct empirically based relationship to performance on HC tasks in the actual noise environments of the CCG and C&P. In addition, the screening tests have adequate and welldefined psychometric properties, i.e., reliability, sensitivity, and validity, so that test results can predict the individual's ability to perform HC tasks in the noise environments, with a known amount of error in prediction. In the proposed method, real-life speech intelligibility performance (or signal detection or sound localization) in different noises is predicted from a screening test in speech intelligibility (or signal detection or sound localization), which does not require any explicit assumption or relationship between auditory tasks.
Research is underway to ensure that the minimum medical standard for hearing is appropriate for the CCG and C&P HC jobs, and that this standard is based on empirical evidence defendable in court. This paper presents the rationale and the detailed method used to establish new hearing criteria for two specific work environments (CCG and C&P), but the approach could be applied to any work environment.
Rationale and design
In order to set new hearing criteria for CCG and C&P departments, the following overall rationale was used [Figure 1]. The research team had to familiarize itself with the seagoing and land-based work activities to identify, in collaboration with job content experts (JCEs), the HC tasks and the locations where these tasks are performed. From the beginning of the project, it was clear that the listening tests could not be performed in the real environments, because of the lack of control over the environment parameters, such as the weather, the wind conditions, the speed of the CCG vessels, etc. In the current project, noise recordings were made in selected locations and recreated in the laboratory where control of noise and signal levels was possible. Neutral test materials (sound stimuli, sentences) were used to limit the effect of experience on performance. The same materials were used in the screening tests, but with reference noises. The ultimate goal was to establish a relation between screening tests and laboratory simulations.
The development of standards for CCG and C&P HC jobs was divided in three distinct stages. [Figure 2] shows an overview of the entire 22month project. Stage 1 (2 months) was a review of literature to select the tests for use as screening measures. Stage 2 (14 months) was the longest part of the project and comprised the noise recordings in the field, data reduction (in collaboration with JCEs), laboratory listening tests with normally-hearing subjects, modeling and determination of performance tables. Stage 3 (6 months) dealt with listening tests with people with different degrees of hearing loss and validation of the model based on normal hearing performance. Finally, in collaboration with CCG and C&P JCEs, minimal acceptable criteria were set and screening scores were proposed.
Overall, forty five normal hearing subjects (Stage 2) and twenty nine subjects presenting various profiles of hearing sensitivity (Stage 3) participated in the project. Six field trips were organized in different regions across Canada (Pacific, Central and Artic, Atlantic, Laurentian and Newfoundland) to cover most of the locations and tasks performed by CCG and C&P employees. The research team worked closely with the JCEs in order to select the most representative environments in which hearingcritical tasks are performed. The following sections present the detailed method used to establish the final criteria.
Development of the approach
Development and validation of the model for predicting functional hearing abilities followed a five-step process. These steps are listed below. In the remainder of this paper we present the methods and results for each of these steps.
1- Identify hearing requirements and measure noise environments during HC jobs;
2- Identify screening measures of functional hearing abilities;
3- Validate the relationship of screening measures to functional hearing abilities in HC noise environments;
4- Relate screening measures to performance in real noise environments for normally-hearing individuals via laboratory studies; and
5- Apply the model to establish functionally-based criteria for HC jobs.
Step 1. Identify hearing requirements and measure noise environments during HC jobs
This step dealt with the identification of the hearing requirements and the description of the environment in which the HC tasks are performed.
Step 1 involves researchers from the University of Ottawa (two audiologists and an engineer) and the House Ear Institute (an experimental psychologist) as well as JCEs from CCG and C&P. [Figure 3] shows the three-level process associated with the identification of hearing requirements and noise environments.
First, JCEs were asked to identify the HC tasks and the workplace locations where these tasks are performed. These tasks comprised sound detection, sound localization, and speech communication; however, in this paper we limit our discussion to the speech communication tasks. The JCEs were also asked to specify performance parameters for each communication task. Parameters included the expected voice level and distance of the communication, and whether repetition of the communication or command was possible.
During a previous Phase (Ritmiller et al., 1999), 34 tasks were identified as being hearing-critical or involving sufficiently frequent auditory demands for Canadian Coast Guard personnel. From these analyses, environments or locations on CCG vessels where these tasks are performed have been identified and summarized. In addition, the Fishery Officer Task Analysis report (Sept. 2001) has been consulted for data on the HC tasks for the fishery officers of the Conservation & Protection seagoing and landbased personnel. This information guided the choice of vessels and environments for the noise measurements. The data collection consisted of recording the entire noise waveform over the measurement session at each chosen location. The measurements were taken in such a way to make it possible to playback (after calibration and equalization) the noise recordings in a laboratory setting so as to reconstruct the noise environments on-board ships or other CCG or C&P workstations.
The research team then went on different vessels, as well as in land and air operations to record the noise environments where CCG and C&P employees perform HC tasks (Laroche et al., 2002). A total of 6 field trips were conducted to collect 112 noise recordings covering all the HC tasks for the CCG and C&P personnel (buoy work, engine monitoring, surveillance, etc.). The measurements at each recording station were stored in digital format onto DAT tapes. A measurement typically consisted of the recording of the noise waveform at a particular location as well as a 30-sec calibration signal before and/or after the noise recording. The length of the noise recordings varied from 1 to 91 minutes depending on the constraints imposed by the environmental conditions or tasks performed.
Two expert panel meetings were held to help the researchers cluster the data set into a comprehensive but manageable set of 15 representative noises from CCG and C&P environments that would be used for the listening tests with human subjects. [Table 1] lists the noise location number (1 to 15) and name, a short description of the CCG and/or C&P environment, the total length of noise records available, and the acoustical characteristics of the noises. The L eq is the equivalent-continuous sound level in dBA and reflects the global level at each location. The difference L 10 -L 90 expresses the temporal fluctuation in the sound level at each location, where L 10 and L 90 are the sound levels in dBA that are exceeded 10% and 90% of the time. The spectral slope is the average slope of the noise spectra at each location in dB/octave. These statistical descriptors take into account the exact weighting of the different noise records within each location as determined by the expert panels. The calculations were derived from the level and one-third octave spectral analyses carried out every 4 seconds for each noise record.
In total, 4 locations (1, 2, 3, 9) were dedicated to fishery officers from the C&P and 8 locations (4, 6, 7, 8, 10, 11, 12, 15) were dedicated to CCG seagoing personnel. In addition, 3 locations (5, 13, 14) were relevant to both CCG and C&P personnel. The 15 locations covered a range of global level, temporal fluctuations and spectral slope. The noisiest locations were 5, 6 and 12, while the least noisy one was location 9. All locations showed a negative average slope (e.g. noise level decreasing with increasing frequency) indicating that the noise was strongest in the low frequency bands. The smallest average slope (flattest spectral distribution) was seen for location 12, and the largest slope (steepest spectral distribution) was seen for location 8. Some locations showed a small difference in L 10 -L 90 , indicating a relatively stable or constant noise environment. Location 3 is a good example of such a location, and to a lesser degree, locations 7, 8, 11 and 15. In contrast, locations with a large difference in L 10 -L 90 indicate a more fluctuating noise environment. This was the case for locations 4, 9, 13 and 14.
For the listening tests, it was necessary to extract a subset of noise samples from the entire population of noise samples at each location. A MATLAB software script was developed to extract a subset of 60 4-sec noise samples for a given location and subject. Special care was taken to ensure that the subset of noises was representative of the noise for a given location. For this purpose, the entire population of noise samples at a location were ordered according to global level in dBA and split into 5 equal bins from the lowest one-fifth of the samples to the highest one-fifth of the samples. Twenty percent of the subset noise samples were taken from each of the 5 bins. Moreover, half the samples extracted from each bin were taken (randomly) among the samples with a spectral slope below the median slope in that bin and the other half were taken (randomly) among the samples above the median spectral slope in the bin. This procedure ensured that each subset of samples would be representative of the range in noise levels and spectral slopes of the entire population of samples at the targeted location. The sampling process was repeated independently for each subject-by-location. Typically, the global level of the 60 samples from the subset was within 0.3 dB of the global level of all the possible samples in a location, and the slope was within 0.1 dB/octave. These noise samples were used in Steps 3 and 4 described below.
Step 2. Identify screening measures of functional hearing ability
The purpose of this step was to review the scientific literature to identify tests that might be used to assess the main auditory demands in the CCG and C&P environments (speech communication, sound localization, sound detection). The research team was also asked to identify screening measures that could be administered under headphones or with loudspeakers in the sound field.
For each auditory demand, the bibliographical search started in databases with keywords such as speech perception, speech intelligibility, sound localization, sound detection, warning sound perception and all their variations. All tests/studies were assessed against specific including/excluding selection criteria. These criteria are listed in [Table 2].
Each test that met the inclusive criteria was analyzed in details and presented in detailed tables in Laroche et al. (2001). From these detailed tables, summary tables were constructed to focus on the most appropriate tests in order to help with final recommendations.
To comply with the project's purpose, the selected tests need to be administered under background noise, involve binaural hearing, and have known psychometric functions. A short test administration time would be considered an added bonus. Most word tests are monaural except for the Virtual Speech Intelligibility in Noise Test (Koehnke & Besing, 1996). Unfortunately, this test does not include established norms and its availability is unknown. Most word tests do not include norms, or, if they do, normative data have only been established in a quiet condition with a relatively small sample of subjects. Indeed, almost all word tests are designed to be administered under quiet conditions.
Several sentence tests allow administration in background noises (e.g., PAL, SPIN, CST, HINT, SIN, QuickSIN). Only a few of these tests allow for binaural listening applications (HINT, SIN & QuickSIN). Unfortunately, the lists of sentence material are not equivalent for SIN (Bentler, 2000). Normative data were obtained under headphones from a relatively small sample for QuickSIN (Killion & Niquette, 2000). The hearing in noise test (HINT) (Nilsson, Soli & Sullivan, 1994) has the most complete published psychometric functions and includes well-established norms. These norms, established with a large number of subjects, were obtained in both quiet and noisy environments, as well as in both free-field protocol and under headphones. The HINT test uses recorded voice and takes about 2 minutes to administer. The HINT can also be used with an adult population, is available in the two official languages at Fisheries and Oceans Canada (English and Canadian French) as well as several other languages. Other suitable tests exist (Plomp & Mimpen, 1979; Hagerman, 1982; Kollmeier & Wesselkamp, 1997), but they are only available in languages other than English (e.g. Dutch, German) and are similar to the HINT. From this analysis, it appears that the HINT is best to evaluate speech intelligibility in noise to predict hearing performances in HC environments.
Step 3. Relate screening measures to performance in real noise environments for normally-hearing individuals via laboratory studies
The overall purpose of this step was to determine the empirical relationship between the screening measures identified in Step 2 and functional hearing ability in real-world noise environments where HC tasks are performed. This relationship takes the form of a statistical model. The model can be used both to characterize the range of functional hearing ability exhibited by normallyhearing individuals in real-world noise environments, and to predict functional hearing ability from scores on the screening measures.
The strategy for accomplishing these purposes is shown graphically in [Figure 4]. Briefly, a sample of 45 normally-hearing subjects was administered the screening protocol. The functional hearing ability of these subjects in laboratory re-creations of 15 different real-world noise environments was also measured. The screening protocol and the functional hearing measures consist of speech intelligibility, sound detection, and sound localization; however, this paper reports only on the speech intelligibility results. The functional measures were normalized by subject and by noise locations, and then pooled. These pooled data were used to create a statistical model of normal functional hearing ability and to predict functional hearing ability for individual subjects in specific realworld noise environments. A detailed description of each step is given below.
The HINT assesses speech intelligibility in noise by adaptively determining the subject's threshold for sentence intelligibility in a stationary, speech spectrum shaped noise presented at 65 dB(A). The HINT was administered both under headphones and in the free field with loudspeakers. Thresholds were measured under three conditions: speech and noise in front at 0° (Noise Front), speech in front and noise at 90° to the right (Noise Right), and speech in front and noise at 90° to the left (Noise Left). For headphone administration, the effects of azimuth on the speech and noise signals reaching the left and right ears were simulated by pre-processing the signals for each ear with generic head-related transfer functions obtained from manikin measurements. Free-field and headphone HINT thresholds are comparable, as long as the free-field measures are not affected by room acoustics (Nilsson et al., 1996; Lamothe et al., 2002). The HINT scores from the three conditions were combined to produce a Composite score according the following formula: Composite = Noise Front + Noise Right/2 + Noise Left/2. The Composite equally weights the contribution of directional hearing, as measured in Noise Right and Left, and nondirectional hearing, as measured in Noise Front. Composite scores, however, may tend to mask the effects of asymmetric hearing losses that produce large differences in the Noise Right and Noise Left thresholds. The screening measures were also comprised of the SAINT sound detection and sound localization tests administered under headphones and in the free field. Results from these measures will be reported separately.
Assessment of functional hearing ability.
Speech intelligibility measures of functional hearing ability were obtained in the laboratory re-creations of real-world noise environments. The 4-sec long noise samples from the field recordings (see Step 1) were reproduced over four loudspeakers in a large double-walled soundproof room. The speakers surrounded the subject at a distance of 1.60 to 1.75 m, and were specially oriented to ensure a diffuse sound field at the subject test position. The speakers were also equalized and adjusted to reproduce accurately the level and spectrum of the noise samples recorded in the field. Lists of HINT sentences were presented from a fifth speaker 1 m in front of the subject at fixed levels to measure percent intelligibility in the noise samples. Sentence presentation levels for a list were chosen to produce expected intelligibility of approximately 40%, 60% or 80%. The average HINT performance-intensity (PI) function for normally hearing individuals was used to specify the presentation levels and resulting S/N ratios at which these levels of expected intelligibility would occur.
The subjects were 45 adult native English speakers with hearing thresholds of 25 dB HL or better in both ears at octave frequencies from 250-8000 Hz. Subjects were grouped into cohorts, and each cohort was tested in different random samples of noise intervals from three different noise environments. Each subject was tested with three lists in each noise environment. Each list was presented at the S/N ratio estimated to produce the three levels of percent intelligibility. This design produces a three-point PI function for each subject in three different noise environments, and three-point PI functions for 9 subjects in each of the 15 noise environments.
Method of analysis.
The data from each noise environment consisted of pairs of S/N ratios and percent intelligibility scores. These data were screened to eliminate outliers by creating bivariate scatterplots for each noise environment that displayed deviations from the median S/N ratio on one axis and deviations from the median percent intelligibility on the other axis. These scatterplots were combined across the 15 noise environments, and the most extreme outliers were identified. Approximately 5% of the data points were eliminated in this fashion. Outliers were assumed to have been due to lapses of attention by the subject (or the experimenter).
After the outliers were removed, the remaining data were submitted to a two-step normalization process. First, the S/N ratios used for intelligibility tests in each noise environment, which had been predicted from the average normative PI function, were normalized by the subject's Composite HINT score. For example, if a subject's Composite HINT score was 3 dB above the Composite norm, 3 dB was subtracted from the S/N ratios used to test that subject in each noise environment. This normalization step corrects the S/N ratios for each subject to the values predicted by the subject's own Composite HINT score. By normalizing the data in this fashion, individual differences in intelligibility among normally-hearing subjects, as determined by the subjects' Composite HINT score, are minimized in the pooled data for each noise environment.
The second normalization was across noise environments. The S/N ratios chosen to produce estimated intelligibility of 40%, 60%, or 80% were comparable across noise environments. However, actual intelligibility scores varied widely across environments, most likely because of differences in the spectral and temporal characteristics of the noise from each environment. The noise levels used to calculate the S/N ratios were based on the L eq of the 4-sec noise samples, which does not capture the shortterm spectral and temporal characteristics of the noise interval. These characteristics were relatively consistent within each noise environment, but varied considerably across environments [Table 1]. The subject-normalized S/N ratios from each noise environment were further normalized by computing a S/N ratio offset for each environment. The S/N ratio offset for each environment was computed by z transforming the percent intelligibility scores from that environment to linearize their relation to S/N ratio. A linear regression function was fit to the intelligibility z-scores in each environment. The intercept of these functions with the S/N ratio axis, which corresponds to a zscore of 0.00 and 50% intelligibility, defined the offset for each noise environment. The offset for each environment was used to shift all of the intelligibility scores for that environment to a normalized S/N ratio of 0 dB.
The intelligibility z-scores were pooled to create a single data set relating subject-normalized and environment-normalized S/N ratio to intelligibility for normally-hearing individuals. The linear regression function predicting mean intelligibility z-scores from normalized S/N ratio was calculated. The linear regression functions corresponding to the 5th and 95th percentiles of the sample were also estimated. These functions were re-expressed as intelligibility scores to yield the standard speech ogive relating intelligibility to S/N ratio. Finally, the offsets for each environment were restored to produce intelligibility ogives positioned at the appropriate S/N ratios for each environment.
The results are presented first for the HINT screening measures and subsequently for the intelligibility tests of functional hearing ability administered in the 15 noise environments. Average headphone HINT scores for the 45 subjects are summarized as follows: Noise Front = -1.8 dB S/N, Noise Right = -9.4 dB S/N, Noise Left = -9.2 dB S/N, Composite = -5.5 dB S/N. Standard deviations ranged from 1-1.4 dB. These results are within 1 dB of the published norms for the English HINT in all conditions. The freefield HINT scores were also quite similar to the headphone scores with the average difference between scores less than 0.4 dB, except for the Noise Right condition where free-field thresholds averaged 2 dB higher (poorer) than headphone thresholds. Free-field thresholds are often elevated in one or both of the noise side conditions because of the presence of acoustically reflective surfaces in the soundroom that affect the directionality of the noise source. The headphone thresholds were used in the remaining analyses.
Speech intelligibility measures of functional hearing.
The intelligibility scores were z-transformed to linearize their relationship to S/N ratio within each noise environment, and linear regression functions were calculated to determine the S/N ratio offset for each environment. [Figure 5] displays intelligibility z-scores as a function of (subject-normalized) S/N ratios for two of the noise environments from [Table 1]: fishing boats with fixed gears (location 1) and fishing boats with mobile gears (location 2). These environments provide examples of data sets that are well fit by the linear regression function (r 2 = 0.81 for fixed gears), and poorly fit (r 2 = 0.26 for mobile gears). The slopes of the two functions are the same (0.16 z/dB), but the intercepts, and thus the S/N ratio offset for the environment, are different (-9.5 dB for location 1, -11.9 dB for location 2). The fit of the linear regression functions varied quite widely across the 15 noise environments. In an attempt to focus the analyses and to make the subsequent validation studies more efficient, the environments were divided into two groups. A total of 8 environments were selected to develop the initial model. This model was then applied to the remaining 7 environments. Likewise, the validation studies were done with only the 8 selected environments. The r 2 values for the 8 selected environments ranged from 0.39 to 0.81, while for the 7 remaining environments these values ranged from 0.02 to 0.47. (Non-statistical sampling requirements of the study design precluded the selection of only those environments with the best linear regression fit for inclusion in the 8 selected environments.)
The S/N ratio offsets were applied to the intelligibility data from the 8 selected environments. Offsets ranged from -7.0 dB S/N for the RHIB/FRC noise environment (location 5) to -14.9 dB S/N for the enclosed bridge and ship's offices (location 13). The magnitude of these offsets can be compared with the Composite (-6.35 dB S/N), and noise front (-2.6 dB S/N) HINT norms, which are based on intelligibility in stationary spectrally-matched reference noise. The HINT noise is intended to produce the most masking of speech per dB of noise. The offsets indicate that the noise from location 5 produced slightly less masking than the reference noise, while the noise from location 13 produced much less masking than the reference noise.
The left panel of [Figure 6] shows the scatterplot of pooled intelligibility z-scores as a function of subject- and environment-normalized S/N ratio for the 8 selected environments. The linear regression function is also shown in this panel. The r 2 variance accounted for by this function is 0.67 (r = 0.82, slope = 0.176 z/dB). This function was transformed with the p transform to produce the traditional intelligibility ogive, as shown in the right panel of [Figure 6]. The middle ogive corresponds to mean predicted intelligibility (50th percentile). The ogives corresponding to the 5th and 95th percentile of the normal sample are also shown. These percentiles are positioned ± 3.5 dB on either side of the mean ogive. When this function is used to predict intelligibility for the entire set of 15 environments, 0.49 of the variance in intelligibility is accounted for (r = 0.70) and the 5 th and 95 th percentiles are positioned ± 4.4 dB on either side of the mean. The loss of predictive accuracy can be traced to 3 of the 15 locations, where the variance accounted for was less than 0.20.
The relative importance of normalizing the S/N ratios with the subjects' Composite HINT score can also be determined by re-calculating the linear regression function for the pooled intelligibility z -scores after normalizing S/N ratios only for noise environment and not for individual differences in Composite HINT scores. The r 2 variance accounted for in this case drops to approximately 0.61 from the original values of 0.67, although the slope of the function (0.178 z /dB) remained essentially the same. This analysis suggests that at least 10% of the variance in functional hearing ability among normally-hearing individuals is due to individual differences measured by HINT.
Step 4. Validate the relationship of screening measures to functional hearing abilities in HC noise environments
The purpose of this step is to validate the statistical model relating S/N ratio to speech intelligibility in real-world noise environments. The validity of the model is defined as its ability to predict the relationship between S/N ratio and intelligibility for a sample of subjects different from the sample used to develop the model, and its ability to make predictions for each noise environment consistent with the communication requirements and performance parameters specified by job content experts (JCEs).
Three different approaches to validation were taken. The first approach addresses the descriptive validity of the model-its ability to characterize the relationship between S/N ratio and intelligibility in a different sample of normally-hearing subjects. The second approach addresses the predictive validity of the model- its ability to predict intelligibility in real-world noise environments for a sample of hearing impaired subjects. This aspect of predictive validation is extremely important if the model is to be used for screening hearing-impaired individuals for HC jobs. This validation involves not only application of the model to a new sample, but also the use of HINT Composite threshold scores, which are likely to be elevated for hearing-impaired individuals, to predict reduced speech intelligibility in real-world noise environments. The third approach may be said to address the convergent validity of the model. The model's predictions of intelligibility for normally-hearing individuals in real-world noise environments are compared with the communication requirements and task performance parameters specified by the JCEs in these same environments. The predictions of the model and the requirements of the JCEs should converge in close agreement if the model is valid.
The model's descriptive validity for normallyhearing subjects was addressed through crossvalidation, since a new sample of normally - hearing subjects was not available. Data from the original sample of 45 subjects tested in the 8 selected noise environments consisted of 186 data pairs of speech intelligibility scores and S/N ratios (after outliers had been removed). Subjects were randomly assigned to two groups with approximately equal numbers of data pairs (94 in group 1, 92 in group 2). Subject-normalized S/N ratios were not changed in the split samples, since these offsets are not sample specific. S/N ratio offsets for each noise environment were recomputed separately for each group, since environment-normalized offsets are sample dependent. New linear regression functions were computed for the subject- and environmentnormalized intelligibility z-scores from each group. The regression function for group 1 could be used to predict intelligibility z-scores for group 2 and vice-versa. If no significant drop in predictive accuracy is observed for either group and both groups exhibit approximately the same regression functions, the validity of the model is supported.
Validation with hearing-impaired individuals.
The predictive validity of the model for hearing impaired individuals was addressed with a new sample comprised of a mixture of 29 normally hearing and hearing-impaired subjects. Subjects were first administered the headphone HINT tests, and their Composite HINT scores were computed. This score, together with environment-specific S/N ratio offsets from the original model, was used to estimate S/N ratios to produce intelligibility scores of 20%, 40%, 60%, and 80% for the 8 selected noise environments. Speech intelligibility was then measured in these noise environments at the predicted S/N ratios. The results, expressed as intelligibility z-scores, were plotted against the intelligibility z-scores predicted by the model. The linear regression function of obtained vs. predicted z-scores and the predictable variance in obtained z-scores was compared with the variance accounted for in the original sample and in the cross-validation samples. Again, if the predictive accuracy for hearing-impaired subjects is comparable to these previous measures, the validity of the predictive model is supported.
Validation from comparisons with JCE communication requirements.
The validity of the model was also addressed by comparisons of the model's predictions of intelligibility in real-world noise environments with the communication requirements established for these environments by the JCEs. Intelligibility predictions were developed using the Pearsons et al. (1977) empirical model to predict expected speech levels in each real-world noise environment. This model predicts conversational speech levels at 1 m as a function of background noise levels. The expected S/N ratios at this distance were computed from the predicted speech levels and the known noise levels. The expected S/N ratios together with the location offsets were used to predict intelligibility for normally-hearing individuals with the current model. The effects of distance on speech level were incorporated by reducing the speech level by 6 dB (half) for every doubling of the distance. Likewise, the effects of loud or shouted levels of speech were incorporated by adding 17 dB to the conversational level for loud speech, and 29 dB for shouted speech. Estimated speech levels in excess of 86 dB(A) 1 m from the talker were not permitted, in accordance with Pearsons' observations that sustained shouted speech rarely exceeds these levels at this distance.
The JCE communication requirements indicated that repetition of spoken communications may be allowed in some HC tasks. A simple means of predicting the improvement in intelligibility provided by repetition was also incorporated in the model. If one assumes that the predicted intelligibility from the model is also the probability, p, of correctly recognizing the sentence, then the probability of incorrect recognition is q = 1- p. If the sentence is repeated and the probabilities of correct and incorrect recognition remain the same and are independent of the initial presentation, the probability of incorrect recognition on the second repetition is also q. Thus, the joint probability of incorrect recognition of a repeated sentence is q 2 and the probability of correct recognition is [1 - q 2 ]. When JCEs specified that repetition were allowed as a HC task parameter, this method of predicting intelligibility was used, together with any corrections for communication distance and voice level.
Cross-validation with normally-hearing subjects.
The linear regression equations predicting intelligibility z -scores from subject- and environment-normalized S/N ratios were almost identical for both groups in the split sample, with a slope of 0.17. The r 2 for the first group was 0.72, as compared with 0.63 for the second group. These values compare favorably with the results for the original pooled sample. The slope of the regression equation in that sample was 0.176 and r 2 was 0.67. It was unnecessary to predict the intelligibility z-scores in group 1 using the regression equation from group two and vice versa, because the two regression equations were almost identical. These results provide strong support for the validity of the model for normally-hearing individuals.
Validation with hearing-impaired subjects.
The new sample for this validation study was comprised of both normally-hearing subjects and hearing-impaired subjects. If normal hearing is defined as four-frequency pure tone averages (PTAs) in both ears less than 30 dB HL, 11 of the 29 subjects had normal hearing. The remaining 18 subjects had four-frequency PTAs in their poorer ear ranging from 30 dB HL to 93 dB HL. More importantly, 11 of the 18 hearing-impaired subjects had asymmetric losses, where asymmetry is defined as a left-right difference in four-frequency PTA of 15 dB or greater. Although this proportion of asymmetric losses may not be representative of the hearingimpaired population, the presence of these subjects in the sample allow the validity of the model to be evaluated for both symmetric and asymmetric audiometric configurations.
The environment-specific and subject-specific S/N ratio offsets and the regression equation for prediction of intelligibility z-scores from the original model was applied to the present sample. The subject-specific offsets were computed from each subject's Composite HINT score. Recall that the S/N ratios used in this study were selected to produce speech intelligibility of 20%, 40%, 60%, and 80% correct. The z-scores for the obtained intelligibility scores were plotted against the zscores for predicted intelligibility scores. If the model accurately predicts intelligibility for hearing-impaired individuals, the common variance shared by the obtained and predicted zscores should be high and the slope of the regression function (in units of obtained intelligibility z-score/predicted intelligibility zscore) should be 1.00, with an intercept of (0,0). The variance accounted for in obtained z-scores was 0.54, as compared to 0.67 in the normal sample. The regression function had a slope of 0.98 and an intercept of (0, 0.18).
When obtained intelligibility was plotted against predicted intelligibility an unusual, but consistent, pattern of results was observed- obtained scores were frequently better but only infrequently poorer than predicted scores. This pattern suggests that the subject-specific S/N ratio offsets, an index of the subject's functional impairment in speech intelligibility, are overestimated by the Composite HINT score. If functional impairment is over-estimated, obtained intelligibility scores would often be better than predicted intelligibility scores- exactly as observed. Recall that a significant portion of hearing-impaired subjects exhibited asymmetric hearing losses. These subjects typically had poor HINT scores in the noise side condition when noise was contralateral to their more impaired ear, relatively better HINT scores for Noise Front where they could rely on their better ear, and relatively better HINT scores for the noise side condition where noise is ipsalateral to their more impaired ear. The Composite HINT score is a weighted combination of front and side HINT scores, so it may over-estimate functional impairment in the Noise Front condition and in one noise side condition for asymmetric losses. The Composite score may also over-estimate functional impairment in diffuse noise conditions, which are analogous to Noise Front HINT, in that directional cues do not differentiate the speech and noise sources.
The above rationale led us to re-calculate subject-specific S/N ratio offsets for the entire sample using the Noise Front rather than the Composite HINT score. The real-world noise environments used in this validation study were diffuse, so offsets based on the Noise Front HINT score were not expected to over-estimate functional impairment for asymmetric losses. When these subject-specific offsets were used, the performance of the model improved markedly. The variance in obtained intelligibility z -scores predicted by the model was 0.61, as compared with 0.54 for the Composite offset. The slope of the function relating predicted and obtained z-scores was 1.01 with an intercept of (0, 0.07). [Figure 7] shows the scatterplot and linear regression equation relating predicted and obtained intelligibility z-scores when the Noise Front HINT score was used to determine the subject-specific offset.
The results using the Noise Front HINT offset support the validity of the predictive model when used with hearing-impaired individuals, including those with asymmetric hearing impairments. The model accounted for 0.67 of the variance in the original sample of normally hearing subjects, and 0.63 to 0.74 of the variance in the cross-validation analysis with normals. The model's variance accounted for with hearing -impaired individuals, 0.61, is not significantly poorer than the lower value from the crossvalidation analysis with normals. Moreover, the slope and intercept for the hearing-impaired sample did not differ significantly from the values predicted under the assumption that the model is valid.
The success of the model when used with the Noise Front HINT offset does not imply that the Noise Front HINT score should be used for screening instead of the Composite HINT score. The better performance of the Noise Front score was due to the use of only diffuse noise environments in the validation study. Real-world noise environments consist of both directional and diffuse noise sources. Moreover, the Composite score, comprised of three independently measured HINT thresholds, will have higher reliability and smaller measurement error than a single HINT threshold.
Validation from comparisons with JCE communication requirements.
The remaining aspect of validity relates to the performance levels in specific real-world noise environments using speech levels and S/N ratios predicted by the Pearsons et al. (1977) model in combination with percent intelligibility for normally hearing individuals predicted by the current model, and their correspondence to the performance levels and parameters (i.e., percent intelligibility, voice level, communication distance, and repetition) specified for these environments by the JCEs. Comparisons of the model predictions and the communication requirements revealed that several of the noise environments required "super normal" speech intelligibility in noise. That is, significant portions of the noise intervals from these environments exhibit noise levels so high that speech produced with the specified parameter will not be intelligible at the specified performance levels. These observations led to reevaluation of both the predictive model and the specified performance levels and parameters.
The outcome of these evaluations was a revised definition of HC tasks by the JCEs: a task is HC only during those times in a real-world noise environment when it can be performed at the specified level of accuracy by normally-hearing individuals using hearing alone. When noise levels are too high to allow the specified accuracy, other senses such as vision or touch are used to augment hearing (e.g., hand signals or physical contact). In these cases, the task is no longer hearing-critical because other sense modalities are used. Alternatively, individuals may wait until the noise level decreases, or move to a different location with a lower noise level, before performing the HC task. In either case, truly HC tasks are only performed in noise environments where normally-hearing individuals can achieve the specified accuracy using only their hearing. The noise statistics for each real-world environment used in the model have been modified to be consistent with this definition. High noise intervals that preclude performance of HC tasks have been eliminated, leaving only those noise intervals in which normals can achieve the performance parameters at all times using only hearing.
[Figure 8] demonstrates the application of these concepts in one of the real-world noise environments, the RHIB/FRC boats (location 5 in [Table 1]). The histogram shows the distribution of background noise levels, ranging from about 60 dB(A) to 95 dB(A) plotted against the left ordinate. Intervals below 75 dB(A) occurred when the motors on the boats were idling or off. Intervals between 75 dB(A) and 90 dB(A) occurred when the motors were running and the boats were underway. The highest interval, 92 dB(A), only occurred when the boats were traveling full speed on the open water. The continuous trace plotted against the right ordinate is the cumulative distribution of the noise levels. Note that low background noise levels (i.e., 65 dB(A) or less) that would allow easy communication occurred less than 10% of the time, while about 50% of the time noise levels were greater than 80 dB(A). The Pearsons et al. (1977) model was used to predict speech levels in each noise interval and the resulting S/N ratios. Percent intelligibility for normally - hearing individuals was predicted from the S/N ratios using the current model, assuming conversational speech levels, a communication distance of 1 m, and no repetitions. Finally, points on the cumulative noise level distribution at which normals would to achieve at least 60%, 70%, 80%, 90% and 100% intelligibility were identified. These points are marked with open symbols and labeled in the figure. Note, for example, that if 100% communication accuracy is specified for HC tasks performed in this environment, noise levels of 67 dB(A) or less are required. These levels occur less than 20% of the time. The JCEs typically specified communication accuracy between 80% and 90% for most HC tasks. Noise levels allowing this range of accuracy occur from 45% to 75% of the time. The initial reaction of JCEs to these values was to question them, but upon further discussion and task analysis they concurred with the predictions of the model. While this evidence of concurrent validity may lack the statistical bases of the cross-validation and predictive validity analyses, it is important nonetheless. For without the concurrence of the JCEs, use of the model could be seriously questioned, regardless of the validity statistics.
Step 5. Apply the model to establish functionally based criteria for HC jobs
The final step consisted of applying the model to establish screening criteria for each task and location. Ultimately, these criteria will be used by the sponsor (Fisheries and Oceans Canada) to identify employees who can safely perform HC jobs.
This step was accomplished in collaboration with JCEs. The screening criteria were expressed as scores on the HINT. When criteria significantly differed from one task or location to the other, minimum criteria were set according to the most demanding task and location. This decision was based on the fact that tasks performed by CCG and C&P employees occur in various locations, and in order to perform at a safe level in all locations, employees need to meet the minimal criteria associated with the most demanding situation. If the criteria associated with certain locations or tasks were more stringent than what would be expected from a normally-hearing group, the normal criteria were instead used. The resulting criteria and rationale were finally validated by the sponsors to insure that they were scientifically based and defendable in court.
Based on computations performed in Step 4, HC tasks and their associated screening Composite HINT scores were calculated. Preliminary screening scores are shown in [Table 3]. For each task and location, the maximum Composite HINT score is shown. Empty cells correspond to locations where no HC tasks are performed. No Composite HINT scores higher than +5 dB or lower than -3 dB are used for screening since limits were imposed on the parameters (distance, voice level, repetition, and performance level) in order to yield realistic screening scores that would not demand performance levels greater than those reached by normally-hearing subjects. The model used to establish screening HINT scores from the task parameters encompassed a range of Composite HINT scores from -10 dB to +5 dB. Recall that the Composite norm was -6.35 dB. This range of scores used in the model exceeds by several dB the range of HINT scores commonly observed not only in occupational health settings, but also in clinical settings.
Locations characterized by high noise levels (e.g., location 8) require better screening HINT scores than locations with medium or low background noise levels (e.g., location 7). As previously stated, the most stringent HINT criteria are to be used for screening only in cases where a recruit or incumbent must perform all HC tasks in all locations. Otherwise, the HINT score can be set at the lowest HINT value specified according to the specific tasks that must be performed and the locations where these occur. For example, an employee required to don lifesaving equipment in all locations would have to obtain a score of -2 dB or better on the HINT screening. But if this same task only occurred in locations other than location 5, the minimal criteria would be set at 1 dB instead of -2 dB.
Such tables could be used to determine the suitability of a new recruit to perform the auditory tasks required by a specific job, or to insure that incumbents still possess the minimal auditory abilities needed to safely perform their job. Failing the screening test does not necessarily mean that the recruit or incumbent cannot perform the job. It could indicate that 1) if no accident/incident has previously occurred, some adaptation process has already taken place (e.g. incumbents with years of experience who use sign language) or 2) adaptation strategies (e.g. hearing aids, noise reduction, tasks modifications) need to be implemented in order to help employees (recruits or incumbents) reach acceptable performance levels. Should the implementation of adaptation strategies prove to be too costly or complex, the employer would be required to prove that all realistically possible solutions were attempted before refusing a candidate.
A new approach to setting screening criteria for HC jobs is proposed. It is well recognized that audiograms are not a valid means of evaluating employees who must perform auditory tasks such as speech perception in noise. This research has examined empirical evidence for use of the Hearing in Noise Test (HINT), a computerized screening measure of functional hearing in reference noises, as an alternative means for making these evaluations. The HINT can be administered quickly over headphones. A sound field version is also available in cases where hearing aids or other devices are worn. Through statistical modeling, performance on this screening measure has been shown to exhibit a predictable relationship to performance of HC jobs and tasks in the actual CCG and C&P noise environments. The HINT also has adequate and well-defined psychometric properties (reliability, sensitivity, and validity) for use in evaluation of individual cases. Knowledge of these psychometric properties insures that predictions of an individual's ability to perform critical auditory skills can be made with a known amount of error in prediction.
When setting screening criteria, several important issues must be taken into consideration. First, many work environments where HC tasks must be performed are quite noisy. For many such environments in the present research even normally-hearing people may not reach performance levels set by the JCEs. Minimal criteria have thus been set for noise levels where communication can effectively take place. The concept of HC tasks was revised throughout this research, since it is incompatible with the fact that these tasks are often performed in high noise levels where normally-hearing people cannot perform. In such situations, individuals necessarily must use other means of communication such as hand signals or touch to augment auditory information. Otherwise, accidents would likely occur, and that has not been the case for jobs with HC tasks in Fisheries and Oceans Canada.
Second, people working in noisy surroundings may suffer from occupational noise-induced hearing loss. These individuals cannot automatically be expected to continually meet the minimal screening criteria throughout their career. Experience and knowledge of the tasks can certainly compensate for the loss of hearing abilities, but in some circumstances, even lengthy experience cannot preclude missing important auditory information in highly critical situations. Again, it appears that other senses (e.g. vision, touch) are used when communication has to be carried on in these particular environments.
Third, workplace adaptations or hearing devices have to be considered when people with hearing loss fail the screening testing. It is only when such adaptations are unrealistic or inappropriate that a recruit or an employee can be declared unfit for a particular HC job.
[Project funded by Fisheries and Oceans Canada, Contract F7053-000009)]
|1||American National Standards Institute (1997). Methods for calculation of the speech intelligibility index. ANSI S3.5-1997. New York: American National Standards Institute.|
|2||Begines, T. (1995). Fitness and Risk Evaluation. Spectrum12: 8-11.|
|3||Bentler, R.A. (2000). List equivalency and test-retest reliability of the Speech in Noise test. American Journal of Audiology, 9:84-100.|
|4||Bherer, L., Deshaies, P., Fortier, P., Laroche, C., Tremblay, V., Turcotte, F. (2002). La surveillance medicale de l'audition exigee par quatre reglements quebecois:Analyse de la pertinence. Avis formule a l'intention du Directeur national de la sante publique. Comite medical provincial en sante au travail du Quebec, 26 p.|
|5||Coles, R.R.A. and Sinclair, A. (1988). Hearing. In Fitness for Work. The Medical Aspects. Edwards, F.C., McCallum, R.I. & Taylor, P.J.(eds). Oxford, Oxford University Press, pp67-89.|
|6||Forshaw, S. and Hamilton, K. (1997). Assessment of occupational hearing requirements. Canadian Acoustics 25: 3-9.|
|7||Forshaw, S., Ritmiller, R., Hodgson, M., Laroche, C., (1999) Estimates of speech intelligibility based on equivalent speech- and noise-spectrum levels and hearing thresholds. Canadian Acoustics 27: 112-113.|
|8||Hagerman, B. (1982). Sentences for testing speech intelligibility in noise. Scandinavian Audiology, 11: 79-87.|
|9||Hetu, R. (1993). Capacites auditives, criteres d'embauche et droits de la personne. Canadian Acoustics 21: 3-14.|
|10||Killion, M. and Niquette, P.A. (2000). What can be puretone audiogram tell us about a patient's SNR loss? Hearing Journal 53: 46-48, 50, 52-53.|
|11||Koehnke J. and Besing JM. (1996). A procedure for testing speech intelligibility in a virtual listening environment. Ear and Hearing 17: 211-7.|
|12||Kollmeier, B. and Wesselkamp, M. (1997). Development an evaluation of a German sentence test for objective and subjective speech intelligibility assessment. Journal of the Acoustical Society of America, 102: 2412-21.|
|13||Lamothe, J., Gascon, C., Lariviere, M., Handfield, M.-F. and Laroche, C. (2002). Normalisation du Hearing in Noise tests (HINT) aupres d'une population francophone bilingue et d'une population Anglophone. Revue d'orthophonie et d'audiologie 26: 81-89.|
|14||Laroche, C. (1994). Cases of possible job discrimination based on hearing loss. Canadian Acoustics 22: 89-90.|
|15||Laroche, C., Giguere, C., Fortin, M. and Soli, S.D. (2001). Hearing Standard for Canadian Coast Guard Seagoing Personnel and CandP Seagoing and Land Based Personnel. Stage 1 Final Report; Literature Review on hearing Tests and detailed Methodology for Stage 2. Submitted to Canadian Coast Guard under Contract No. F7053-000009, (27 July 2001), 16 p. (+Appendices).|
|16||Laroche, C., Giguere, C., Soli, S. (2002). Development and validation of hearing standards for Canadian Coast Guard seagoing personnel and CandP and land-based personnel, Phase II. Report on Stage 2. Pilot study and validation. Submitted to Canadian Coast Guard under Contract No. F7053-000009, (31 July 2002), 33 p. (+Appendices).|
|17||MacLean, S. (1995). Employment Criteria in Hearing Critical Jobs. Spectrum12: 20-23.|
|18||Nilsson, M., Soli, S.D., Sullivan, J.A. (1994). Development of the Hearing in Noise Test for the measurement of speech reception thresholds in quiet and in noise. Journal of the Acoustical Society of America 95: 1085-1099.|
|19||Nilsson, M., Soli, S.D., Sumida, A. (1996). A definition of normal binaural sentence recognition in quiet and noise. House Ear Institute, pp1-13.|
|20||Pavlovic, C.V., Studebaker, G. A. and Sherbecoe, R.L. (1986). An articulation index based procedure for predicting the speech recognition performance of hearingimpaired individuals. Journal of the Acoustical Society of America 80: 50-57.|
|21||Pearsons, K.S., Bennett, R.L. and Fidell, S. (1977). Speech Levels in Various Noise Environments. EPA 600/1-77-025, US Environmental Protection Agency.|
|22||Plomp, R. and Mimpen, A.M. (1979). Improving the reliability of testing the speech reception threshold for sentences. Audiology, 18:43-52.|
|23||Ritmiller, L., Forshaw, S., Hodgson, M., Laroche, C., (1999) Development of bona fide occupational requirements for hearing in Canadian Coast Guard operations. Canadian Acoustics, 27:108-109.|
|24||Smoorenburg (1992). Speech reception in quiet and in noisy conditions by individuals with noise-induced hearing loss in relation to their tone audiogram. Journal of the Acoustical Society of America 91: 421-437.|
|25||Soli, S.D. and Vermiglio, A. (1999). Assessment of functional hearing abilities for hearing critical jobs in law enforcement. Report for the California Peace Officers Standards and Training Commission, 29 pp.|
|26||Weeks, J.L., Levy, B.S. and Wagner, G.R. (Eds.) (1991) Preventing Occupational Disease and Injury. American Public Health Association, Washington, D.C.|
|27||WHO (2000). Guidelines for Community Noise. Berglund, B., Lindvall, T., Schwela, D., Goh, K.-T. (eds.) World Health Organization, Geneva.|