International Journal of Audiology

ISSN: 1499-2027 (Print) 1708-8186 (Online) Journal homepage: http://www.tandfonline.com/loi/iija20

Do you hear the noise? The German matrix sentence test with a fixed noise level in subjects with normal hearing and hearing impairment Nina Wardenga, Cornelia Batsoulis, Kirsten C. Wagener, Thomas Brand, Thomas Lenarz & Hannes Maier To cite this article: Nina Wardenga, Cornelia Batsoulis, Kirsten C. Wagener, Thomas Brand, Thomas Lenarz & Hannes Maier (2015) Do you hear the noise? The German matrix sentence test with a fixed noise level in subjects with normal hearing and hearing impairment, International Journal of Audiology, 54:sup2, 71-79, DOI: 10.3109/14992027.2015.1079929 To link to this article: http://dx.doi.org/10.3109/14992027.2015.1079929

Published online: 10 Nov 2015.

Submit your article to this journal

Article views: 81

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=iija20 Download by: [Purdue University Libraries]

Date: 15 February 2016, At: 04:04

International Journal of Audiology 2015; 54: 71–79

Original Article

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

Do you hear the noise? The German matrix sentence test with a fixed noise level in subjects with normal hearing and hearing impairment Nina Wardenga†, Cornelia Batsoulis†,§, Kirsten C. Wagener*,‡, Thomas Brand*,#, Thomas Lenarz*,† & Hannes Maier*,† *Cluster of Excellence ‘Hearing4all’, Hannover & Oldenburg, Germany, †Department of Otolaryngology, Hannover Medical School, Hannover, Germany, ‡Hörzentrum Oldenburg GmbH, Oldenburg, Germany #Department of Medical Physics, Carl-von-Ossietzky-University Oldenburg, Oldenburg, Germany, §MED-EL Medical Electronics, Hannover, Germany

Abstract Objective: The aim of this study was to determine the relationship between hearing loss and speech reception threshold (SRT) in a fixed noise condition using the German Oldenburg sentence test (OLSA). Design: After training with two easily-audible lists of the OLSA, SRTs were determined monaurally with headphones at a fixed noise level of 65 dB SPL using a standard adaptive procedure, converging to 50% speech intelligibility. Study sample: Data was obtained from 315 ears of 177 subjects with hearing losses ranging from  5 to 90 dB HL pure-tone average (PTA, 0.5, 1, 2, 3 kHz). Results: Two domains were identified with a linear dependence of SRT on PTA. The SRT increased with a slope of 0.094  0.006 dB SNR/dB HL (standard deviation (SD) of residuals  1.17 dB) for PTAs  47 dB HL and with a slope of 0.811  0.049 dB SNR/dB HL (SD of residuals  5.54 dB) for higher PTAs. Conclusion: The OLSA can be applied to subjects with a wide range of hearing losses. With 65 dB SPL fixed noise presentation level the SRT is determined by listening in noise for PTAs  ∼47 dB HL, and above it is determined by listening in quiet.

Key Words:  Speech reception threshold; matrix sentence test; German sentence test (OLSA) in noise; medical audiology; discriminative power

­ or quantifying speech intelligibility of individuals in different F situations, a broad variety of tests is currently available. They differ not only in terms of the speech material (e.g. logatomes, monosyllables, numbers, and meaningful or meaningless sentences), but also in their mode of application (e.g. presentation in quiet or in noise, fixed or adaptive levels of speech and/or noise level, and type of noise). The most common German speech test, used by the majority of hospitals, medical practitioners, and hearing-aid dispensers, is the ʽFreiburger Sprachtestʼ (Freiburg speech test) (Hahlbrock, 1953). Despite its popularity, it remains subject to criticisms, including differences between test lists, the limited number of available lists, the use of out-dated words, and most importantly, the lack of the possibility to determine speech intelligibility in noise (Kießling, 2000). Testing whole sentences and intelligibility in noise allows a more adequate simulation of communication in everyday situations (Thiele et al, 2011; Zokoll et al, 2013; Akeroyd et al, 2015; Kollmeier et al, 2015) and can give additional information about the actual hearing impairment and the benefit of hearing aids and implants. Several sentence tests in the German language are available, such as the

Oldenburg sentence test (Wagener et al, 1999a,b,c; Wagener & Brand, 2005), the Göttingen stentence test (Kollmeier & Wesselkamp, 1997), the Hochmair-Schulz-Moser sentence test (Hochmair et  al, 1997), and the Basel sentence test (Tschopp & Ingold, 1992); which are all implemented in noise. These tests have been used for measuring the benefit and status of rehabilitation for patients with hearing aids and implants (Müller-Deile, 2009; Hey et al, 2014; Kollmeier et al, 2011; Kleine Punte & Van de Heyning, 2013). The Oldenburg sentence test (OLSA) was developed and evaluated for testing speech intelligibility in noise (Wagener et al, 1999a,b,c; Wagener, 2004) and is also applicable to quiet conditions (Wagener & Kollmeier, 2004). It originates from a test by Hagermann (1982) who introduced a Swedish sentence test based on a 5  10 word matrix with columns containing ten names, verbs, numbers, adjectives and objects (see Table 1). By randomly selecting words from each column, syntactically fixed sentences of semantically low predictability, e.g. ʽTanja malt neun nasse Messerʼ (Tanja paints nine wet knives), are generated. Each word of the matrix is used only once per list of ten sentences and two lists of ten sentences are

Correspondence: Hannes Maier Department of Otorhinolaryngology and German Hearing Centre Hannover, Hannover Medical School, Carl-Neuberg-Str. 1, 30625 Hannover, Germany.E-mail: [email protected] (Received 20 Dec 2014; accepted 2 Aug 2015) ISSN 1499-2027 print/ISSN 1708-8186 online © 2015 British Society of Audiology, International Society of Audiology, and Nordic Audiological Society DOI: 10.3109/14992027.2015.1079929

72     N. Wardenga et al.

Abbreviations

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

AC ANOVA dB HL MV NH OLSA PTA SD SII SNR SPL SRT

Air conduction Analysis of variance Decibel Hearing level Mean value Normal-hearing Oldenburg sentence test Pure tone average Standard deviation Speech intelligibility index Signal to noise ratio Sound pressure level Speech reception threshold

usually combined to increase accuracy. The phoneme distribution of the speech material approximates the German language (Wagener et al, 1999a). In contrast to everyday phrases, the sentences generated from the matrix have low semantical redundancy and cannot be predicted based on the sentence context. Because of the semantically unpredictable structure, the lists cannot be memorized easily and thus, may be used repeatedly. Different general approaches exist in the execution of such tests (Wagener & Brand, 2005; Wagener et  al, 2006). For example, the OLSA can adaptively determine the signal to noise ratio where 50% of words is understood (commonly called speech-reception threshold (SRT)) using an iterative adaptation of the speech presentation level in the presence of a constant noise level. One of the reasons that no speech test in noise is prevalent in German clinical practice may be due to the fact that reference data is available mostly for normal-hearing (NH) listeners and that a higher degree of experience is required to construe measurement results. This study should support the interpretation of results by presentTable 1. Word matrix of the German version of the OLSA matrix sentence test. By combining words from each column, random sentences like the example (highlighted in bold, ʽBritta buys twelve pretty cansʼ) are created. Name Peter Peter Kerstin Kerstin Tanja Tanja Ulrich Ulrich Britta Britta Wolfgang Wolfgang Stefan Stefan Thomas Thomas Doris Doris Nina Nina

Verb

Number

Adjectiv

Object

bekommt gets sieht sees kauft buys gibt gives schenkt donates verleiht lends hat has gewann wons nahm took malt paints

drei three neun nine sieben seven acht eight vier four fünf five zwei two achtzehn eighteen zwölf twelve elf eleven

große big kleine little alte old nasse wet schwere heavy grüne green teure expensive schöne pretty rote red weiße white

Blumen Flowers Tassen cups Autos cars Bilder pictures Dosen cans Sessel chairs Messer knives Schuhe shoes Steine stones Ringe rings

ing reference data for listeners with different degrees of hearing loss. We adaptively measured the SRT with the OLSA at a fixed noise level of 65 decibel (dB) sound pressure level (SPL) in a broad range of individuals with normal and with impaired hearing in an unaided situation. This provides well defined acoustical boundary conditions and the result of the measurement describes the communication abilities of the subject at this moderate noise level. These results were correlated with the degree of hearing loss in terms of pure-tone averages (PTAs) derived from air conduction (AC) puretone audiometric thresholds at 0.5, 1, 2, and 3 kHz and are compared to predictions using the speech intelligibility index (SII). The use of a fixed noise level of 65 dB SPL, which is commonly used in audiological and clinical practice, regardless of the subject’s hearing loss, is different from other approaches where the noise level was individually chosen according to the subject’s medium loudness level. The approach presented here with constant fixed noise presentation level for all subjects allows for comparing the communication abilities due to speech intelligibility of different hearing losses in a given communication environment. When using a common fixed noise level the masking effect of the noise will disappear beyond a certain hearing loss as the noise is no longer audible to the subject. Consequently, subjects with normal hearing or moderate hearing loss will be tested in a typical noise condition, whereas subjects with a higher degree of hearing loss will be tested effectively in a hybrid or in-quiet condition. The other approach with individually chosen noise presentation levels would allow for determining the individual speech intelligibility in noise which can be considered as one functional component of the individual communication ability. Besides the aim of this study to provide comparison data for the presented approach that is relevant in clinical practice, another goal was to quantify the critical hearing loss (where the noise becomes irrelevant) in terms of PTA for a fixed noise presentation level of 65 dB SPL. Although, in routine clinical practice it is of greatest importance for any noise level used that an upper limit of the validity1 of the SRT in noise can be easily determined, we wanted to investigate an exemplary, commonly used configuration with a fixed noise presentation level. As an additional aim we investigated the possibility of estimating the SRT in noise from the hearing loss. For this purpose the PTA across the hearing thresholds in dB hearing level (HL) at 0.5, 1, 2, and 3 kHz was used and compared to a SII based prediction model. The PTA is a single number measure characterizing the hearing loss. It is widely used and recommended to characterize hearing losses in comparison to speech intelligibility in quiet (Gurgel et al, 2012). Furthermore, the PTA has the advantage that it requires no elaborate calculation and can be determined easily from a given audiogram even by a medical practitioner. The SII (ANSI S3.5-1997) assumes that the speech intelligibility is determined by the maximum of masking noise and hearing threshold within each auditory frequency channel. In other words, the hearing loss is handled as an internal noise. The SII requires more calculations than the PTA. However, even though very simple procedures to calculate a slightly simplified version of the SII are available (Mueller & Killion, 1990), they have not been used in routine clinical practice. For the comparison of PTA and SII as possible predictors for the SRT in noise, it has to be taken into account that the SII neglects the hearing threshold completely in all frequency regions in which the masking noise is audible. By definition, this neglects the possible consequences of auditory neuropathy, where reduced speech processing occurs even at clearly audible levels despite only a low level hearing

OLSA in unaided subjects with hearing loss      73

loss. In this study the PTA was investigated because this very simple approach might be advantageous compared to the SII since it takes low level hearing losses into account. Hence, one hypothesis of this study was that the PTA is able to explain some variance observed in SRTs in noise for low degrees of hearing loss that cannot be explained by the SII.

Methods

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

Subjects and demographics In this study, 177 subjects (96 males and 81 females) participated. The average age at the time of examination was 54.6 years (ranging from 17 to 82 years). Subjects with different degrees of hearing loss from our clinical routine were asked to volunteer in additional tests required for this study. The test battery used for analysis consisted of the following items administered monaurally with headphones (Sennheiser, HDA200): 1. Air conduction pure-tone audiogram at 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, and 8 kHz. 2. The matrix sentence test ʽOldenburger Satztestʼ (Oldenburg sentence test, OLSA) at 65 dB SPL noise level. Tests and the following analysis were performed with monaural unaided ears. Most subjects contributed data for both ears (315/177; ears/participants).

Categorization of hearing impairment groups In our approach to quantify the PTA limit where the SRT in noise is determined we tried to use as few as possible a priori assumptions and to iterate this limit in a stepwise procedure, starting with a coarse categorization of hearing losses. Based on the pure-tone audiogram data, single ears were categorized into five separate groups: one group with normal hearing, three groups with different degrees of hearing impairment according to their respective PTA (average hearing level at 0.5, 1, 2, and 3 kHz, according to the guideline of Gurgel et  al (2012)), and one group with special characteristics in the audiogram. The groups are defined as follows: Group A: Normal-hearing, AC thresholds at 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, and 8 kHz  20 dB HL (one exception at a single frequency with a threshold equal to 20 dB HL was permitted). Group B: Mild hearing loss, not in group A and PTA  40 dB HL. Group C: Moderate hearing loss, 40 dB HL  PTA  60 dB HL. Group D: Severe hearing loss, PTA  60 dB HL. Group E: Special cases, fulfilling one of the following two conditions: E1: Air conduction thresholds for the determination of the PTA at 3 kHz  110 dB HL (audiometer limit). Missing values were set to 115 dB HL for the calculation of the average PTA. E2: Subjects with a steep hearing loss: ears having a difference in thresholds at 0.25 and 3 kHz  80 dB HL.

were calibrated annually in accordance with international standards (EN 60645-1/-2/-4, EN ISO 389-1/-2/-3/-4/-5/-7, ISO 389-8). The OLSA was executed in a sound-proofed chamber. The sound materials were presented with the PC-based Oldenburg Measurement Application software (Hörtech GmbH, OMA) monaurally, unaided via headphones (Sennheiser, HDA200). The headphones were freefield equalized according to the international standard (ISO 389-8) and the measurement setup was calibrated in dB SPL as specified by the manufacturer (Hörtech GmbH).

Procedure Pure-tone thresholds were measured at 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, and 8 kHz, with frequencies being presented in randomized order to determine the degree of hearing loss and to classify the subjects’ ears to subgroups. Each frequency was repeated until the indicated thresholds were consistent. In cases where the audiometer limit of 110 dB HL was exceeded, missing values were estimated for statistical analysis using the audiometer limit plus 5 dB. The OLSA was performed in the open set presentation mode, where the subject was instructed to verbally repeat all words from a presented sentence. Each correctly repeated word was recorded and scored (word-scoring) by the test instructor on a touch screen. According to the number of correctly understood words in the preceding sentence, the speech level was adapted for the next sentence by a word score dependent increment in the software. Starting at 0 dB signal-to-noise-ratio (SNR) and a fixed noise level of 65 dB SPL, the procedure iterates an estimate of the presentation level where 50% of presented words are understood. The estimation of the SRT is determined by a maximum likelihood procedure (Brand & Kollmeier, 2002). In normal-hearing subjects the reference SRT is  7.1   0.2 dB SNR at 65 dB SPL noise level when speech and noise are presented monaurally via headphones. The stationary speech simulating noise used (olnoise, Wagener et al, 1999c) was generated by using the entire German matrix material and has the same long-term average spectrum (Wagener et al, 1999a). Noise presentations started and ended 500 ms before and after each sentence. To receive reliable and reproducible results, subjects must be acquainted with the speech material and the structure. To reduce training effects (Wagener et al, 1999c) and to become familiar with the speech material, two well recognizable lists were presented binaurally at a fixed level. The first training list with 20 sentences was presented in quiet at usually 65 dB SPL, whereas the second training list was presented at the same presentation level in combination with the noise presented at 55 dB SPL. In profoundly deaf subjects, the level of the speech material was increased to a comfortable level up to a maximum of 95 dB SPL to ensure maximum understanding during the training phase. After the training, the SNR was determined for each ear separately with the OLSA presented to the listener’s better ear first (determined by PTA). The noise level remained constant at 65 dB SPL and the signal adaptation procedure started at 65 dB SPL signal level (0 dB SNR). All subjects completed the measurement under the same conditions, independent of their individual threshold.

Speech intelligibility index Measurement setup AC thresholds were measured monaurally, unaided via headphones (Sennheiser, HDA200) in a sound-proofed chamber using a PCbased audiometer (Audio-DATA, AD2017 or AD17). The devices

The SII is a standard procedure to predict the intelligibility of speech in quiet as well as in noise conditions. The method goes back to the work of French and Steinberg (1947) as well as Fletcher and Galt (1950). The method was revised several times. In this study the ANSI

a one-way analysis of variance (ANOVA) with post hoc Bonferroni procedure was used.

standard S3.5-1997 (which is still valid) was used. Basically, the SII is a weighted sum of SNRs (limited to values between  15 and  15 dB) in different frequency bands. The hearing loss in each frequency band is taken into account by an equivalent internal noise spectrum level. The equivalent disturbance spectrum level used for the SNR calculation is set to the maxima of the equivalent internal noise spectrum level and the external noise spectrum level. The calculation also includes some non-linearities as at higher levels a level distortion factor reduces the SII result and as self-masking of speech is taken into account. The SII was implemented according to the one-third octave band SII procedure (Table 3 of ANSI S3.5-1997). Identical frequency spectra for speech and noise (according to the frequency spectrum of the olnoise) were used for the calculation. The frequency weighting was done according to the band importance function for speech in noise (ANSI S3.5-1997, Table B2, SPIN). Using these parameters the SII threshold value that is related to the reference SRT of  7.1 dB SNR for subjects with normal hearing is 0.227. For each ear measured in this study the minimum SNR that yields this SII threshold was calculated using a step size of 0.1 dB. This SNR was used as the predicted SRT.

Results Ears included in this study were categorized according to the audiogram criteria as normal hearing ears (group A, N  55, PTA 1.8  4.6 dB HL, mean value (MV)  standard deviation (SD)), mild hearing loss (group B, N  102, PTA 24.2  10.4 dB HL), moderate hearing loss (group C, N  77, PTA 49.6  5.9 dB HL), severe hearing loss (group D, N  66, PTA 70.0  7.5 dB HL), and exceptions (group E, N  15). The distributions of air conduction pure-tone thresholds of all groups are shown in Figure 1. The hearing loss of subjects in group A–E was not independent of age. The analysis of the age distributions (Kolmogorov-Smirnov test, p  0.05) indicated that a normal distribution of age in all groups except group E (52.3  14.4 years (MV  SD)) can be assumed. The subjects in group A (30.8  7.1 years) were significantly younger (one-way ANOVA, Bonferroni, p  0.001) than in all other groups, whereas subjects in group B (55.1  12.5 years) were significantly younger (p  0.01) than subjects in group C (61.7  14.2 years) and D (62.4  12.1 years). All other differences in age between groups were not significant. Figure 2 shows the mean values and the standard deviations of measured SRTs across subjects in the groups A–E using the OLSA. Within all groups the assumption of a normal distribution of SRTs obtained with the OLSA was verified. SRTs of groups (A–D) were statistically different from each other (one-factorial ANOVA

Statistics For all statistical analyses, the Software IBM SPSS Statistics 22 was used. To test data for normal distribution the Kolmogorov-Smirnov test and the Shapiro-Wilk test were used. To test for homogeneity of variance the Levene’s test was applied. The relationship between the SRT and the degree of hearing loss was determined by using Pearsonʼs linear regression. To test for differences in the age of subjects in group A–E,

Hearing threshold [dB HL]

0.5

1

2

34

Frequency [kHz]

Frequency [kHz]

Frequency [kHz] 0.25

6 8

0.25

0.5

1

2

34

6 8

0.25

0.5

1

2

34

6 8

0 20 40 60 80

100

Hearing threshold [dB HL]

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

74     N. Wardenga et al.

(A)

(B)

(C)

(D)

(E1)

(E2)

0 20 40 60 80

100

Figure 1. Average and distribution (median, 10th, and 90th percentiles) of the air conduction pure-tone thresholds at 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, and 8 kHz for all subgroups.

OLSA in unaided subjects with hearing loss      75



the ability to compensate noise monotonically decreases with hearing loss. The distribution is separated into at least two ranges that can be described by different slopes. If the range of analysed hearing loss is restricted to normal hearing (group A) and mildly-impaired subjects (group B), a linear approximation can be used to characterize the data in this range resulting in a highly significant correlation coefficient (r²  0.569, p  0.001 Pearson correlation coefficient) of the linear approximation (Figure 3, middle panel). Therefore, the linear dependence of the SRT [dB SNR] on the PTA [dB HL] in group A and B can be given by the relationship:

30

10

better performance

SRT [dB SNR]

20

0

SRTPTA  40 dB HL  (0.093  0.006)  PTA  (6.135  0.139) (1a) The slope of ∼ 1 dB SNR / 10 dB HL further implies that the ability to discriminate speech in noise is intimately linked to the loss in threshold, if the PTA does not exceed 40 dB HL. When analysing the data of group D (severe hearing loss, PTA  60 dB HL) a different significant linear correlation with the PTA was found:

–10

Figure 2. Adaptive mean SRTs determined with the OLSA at a fixed noise level of 65 dB SPL for all groups. Circles depict mean values, and standard deviations are indicated by whiskers.

SRTPTA  60 dB HL  (0.856  0.104)  PTA  (42.462  7.354) (2a)

(Games-Howell test), p  0.001); whereas group E overlaps with some of the other groups (Kolmogorov-Smirnov, p  0.05). The dependency of monaurally determined SRTs in 65 dB noise on the PTAs is plotted in Figure 3. As can be seen in the upper panel,

Here the one-tailed Pearson correlation coefficient shows a significant linear correlation between the PTA and the SNR (r2  0.512, p  0.001), as illustrated in the lower panel of Figure 3.

better performance

35 30 25

0

group A group B group C group D group E1 group E2

group A group B

–2

20

better performance

15 10 5 0 –5 –10

SRT [dB SNR]

40

SRT [dB SNR]

–4 –6 Reference in olnoise

–8 –10

0

10

20

30

40 50 60 70 PTA [dB HL]

40

80

90 100 110

–10

0

10 20 PTA [dB HL]

30

40

group D

35 30

SRT [dB SNR]

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

group A group B group C group D group E1 group E2

25 20 15 10 5 0 60

65

75 70 PTA [dB HL]

80

85

Figure 3. SRTs determined with the OLSA at 65 dB SPL noise level as function of PTA (0.5–3 kHz). Upper panel: All groups A E. Middle panel: Detailed view of groups A and B. The hatched range shows the reference SRT range for normal-hearing subjects at 65 dB SPL noise. Lower panel: Detailed view of group D. Linear regressions (1a) and (2a) of the respective subgroups are shown as solid black lines, and the 95% confidence interval as dotted lines.

The upper panel in Figure 3 (see also Figure 5) indicates that group E2 might be better described by the linear approximation (1a) than by the linear regression relevant for this PTA range (2a). The hearing losses of group E2 are characterized by a steep hearing loss of  80 dB HL between 0.25 and 3 kHz and a good residual hearing at low frequencies ( 30 dB HL; 0.25 kHz). The PTAs of group E1 have missing pure-tone data (mostly of 3 kHz). The PTA is a best case estimated by replacing non-measurable thresholds by the audiometer limit plus an increment. Even if the linear regression coefficient of (1a) was increased by combining groups A, B, and E, group E1 was treated as an exception that may underlie parameters that cannot be assessed by the PTA (e.g. duration or onset of hearing loss) and was excluded from the following analysis. Although groups A, B, and D show a clear linear correlation between the PTA and the SRT, group C visually shows a transition between these two analysed domains and required a more detailed a posteriori analysis.

A posteriori analysis The linear regressions (1a) and (2a) result in an intersection of both linear domains at a PTA of 47.6 dB HL. At this point the slope and spread of the SRTs changes significantly. Therefore, group C can be segmented into two hypothetically different domains with C1   47.6 dB HL and C2  47.6 dB HL. In the original ranges AB (PTA  40 dB HL) and D (PTA  60 dB HL) as well as in the sub-ranges C1 (40  PTA  47.6 dB HL) and C2 (47.6  PTA   60 dB HL) the differences ʽDeltaʼ of the linear regressions (1a) and (2a) to the measured SRTs in the respective ranges was computed. The found residues Delta were normally distributed (KolmogorovSmirnov test and Shapiro-Wilk test) in all groups. The standard deviations in the range AB (SD  1.12 dB) were found to be very similar to C1 (SD  1.48 dB) and those in D (SD  6.28 dB) were very similar to C2 (SD  4.46 dB). This in combination with the pronounced change in SD at the separation point at PTA  47.6 dB HL justifies the assumption that sub-group C1 belongs to group AB and sub-group C2 belongs to group D. 30

The analysis of the residues Delta in the newly created ranges ABC1 and DC2 demonstrates that they are normally distributed (KolmogorovSmirnov test and Shapiro-Wilk test) with SD smaller or close to the original groups (ABC1, SD  1.17 dB; DC2, SD  5.54 dB). Figure 4 displays the histogram of the residues Delta in the initial ranges AB and D compared to the newly created ranges ABC1 and DC2. The linear regression (1b) of the data in the new ranges ABC1 leads to an increase in the already highly significant correlation coefficient (r²  0.617, p  0.001 Pearson correlation coefficient) of the linear approximation: SRTPTA  47.6 dB HL  (0.094  0.006)  PTA  (6.159  0.141) (1b) Also the linear regression (2b) of the data in the new range DC2 leads to an increase correlation coefficient (r²  0.702, p  0.001 Pearson correlation coefficient). SRTPTA  47.6 dB HL  (0.811  0.049)  PTA  (39.407  3.107) (2b) The linear regressions (1b) and (2b) are illustrated in Figure 5. When the intersection of the linear approximations (1b) and (2b) is calculated, the intersection is marginally reduced to a PTA of 46.4 dB HL. The difference to the originally used threshold of 47.6 dB HL to dichotomize the range was deemed within the accuracy of the analysis and a further iteration of the a posteriori analysis was not performed.

Discussion This study presents SRT data of N  315 measured ears under standardized conditions in a mono-centre study, that covers a range of PTAs from 0 up to 85 dB HL. The specific condition of this study, using a fixed noise level of 65 dB SPL, assessed the individual communication abilities of subjects in a well-defined reference communication situation. The SRTs in noise determined in this study can be separated into two PTA domains with a transition at a PTA of approximately 47 dB HL. In both domains linear regressions were highly significant and the residues were normally distributed. The SRT of subjects with normal hearing and subjects with a hearing loss below that margin (groups A, B, C1) increased only slightly with increasing PTA with a slope of 0.094 dB SNR/dB HL 14

ABC1 AB

10

Frequency

20

15

10

8

6 4

5

0

DC2 D

12

25

Frequency

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

76     N. Wardenga et al.

2

–4

–2

0 DELTA

2

4

0

–15

–10

-5

0

5

10

-15

DELTA

Figure 4. Histogram of the differences between the observed data and the linear regression. All groups can be assumed normally distributed (K-S test and Shapiro-Wilk test) with the standard deviations: AB, SD  1.12 dB; D, SD  6.28 dB; ABC1, SD  1.17 dB; DC2, SD  5.54 dB).

OLSA in unaided subjects with hearing loss      77

better performance

30

SRT [dB SNR]

20

105

group ABC1 group DC2 group E1 group E2

95 85

10

75

0

65

–10

55

–20

45

–30

35

–40

25 Reference in quiet

–50 10

20

30

40 50 60 70 PTA [dB HL]

80

15

90 100 110

Figure 5. SRTs determined with the OLSA at 65 dB SPL noise level as function of the PTA (0.5–3 kHz). The solid lines show the linear regressions in black of group ABC1 and in grey of group DC2 and the light dashed lines depict the 95% confidence intervals of the linear regressions in the respective ranges. The hatching shows the range for normal-hearing subjects in quiet.

this study. Glyde et  al (2013) demonstrate a linear relationship in hearing-impaired subjects up to a PTA of 60 dB HL. Although these authors have a focus on spatial processing in noise tested with the ʽListening in spatialized noise sentences testʼ (LiSN-S, Cameron et al, 2009) they found a linear relationship with a comparable slope (∼0.09 dB SNR/dB HL) even in the aided condition using a NAL-RP prescribed insertion gain if the speaker and a competing talker with the same voices are presented from 0°. Beside the linear dependence found in the lower PTA range (group A, B, C1) the narrow distribution of the residues gives rise to the possibility that the SRT in noise can simply be estimated with usable accuracy from the PTA. After exclusion of the few exceptions according to the rules defined for group E, the expected SRT can be estimated with a precision of  2.3 dB (95% range) without testing the SRT in noise directly. As the SII is a reference method for estimating intelligibility in noise, we compared the SII predictions to the PTA based predictions and to the measured results (Figure 6). For subjects in groups A and B the SRTs were estimated 1.0 dB lower (ʽbetterʼ) by the SII than actually measured, whereas for subjects in group C and in group D the SRTs were predicted 3.1 dB and 6.6 dB worse than measured, respectively. Interestingly, at least some subjects in our groups E1,2 were outliers and had a much better performance than predicted by better performance > 45 45 35

group A group B group C group D group E1 group E2

30 25 20

better performance

SRTpredicted [dB SNR]

40

15 10 5 0 –5 –10 0

10 20 30 40 50 60 70 80 90 100 110 PTA [dB HL]

better performance 40 35 30 25

group A worse than predicted group B group C group D group E1 group E2

20 15

better performance

SRTmeasured [dB SNR]

(1b). Despite the broad range of hearing losses in that range, a linear approximation described the data very well with normally distributed residuals (SD  1.17 dB). Above this margin of a PTA of 47 dB HL the cohort was better described by a linear function with a much higher slope of 0.811 dB SNR/dB HL (2b). Also, the distribution of the residuals (SD   5.54 dB) was increased by a factor of five in the upper range, being an additional indicator for the transition between ranges. As expected the masker with a level of 65 dB SPL was not perceivable in the group with the most pronounced hearing loss and SRTs determined in noise were in more or less quiet or in inaudible noise, respectively. The hypothesis that the regression of the upper PTA range (2b) describes the hearing in quiet is supported by extrapolating the (2b) regression. Despite the huge gap in PTA between the data and the extrapolation target PTA value of 0 dB HL, the extrapolated SRT of hypothetical normal-hearing subjects in the absence of noise is 25.6 dB SPL. This is only approximately 6 dB higher than the reference SRTs experimentally found for NH listeners (Wagener, 2004, Wittkop et  al, 2011). On the other hand the refined linear regression (1b) for low hearing losses predicts an average SRT of  6.2  1.2 dB SNR for ʽNHʼ subjects with a PTA  0 dB HL. This is approximately 1 dB SNR higher (ʽworseʼ) than the reference group (7.1  0.16 dB SNR) of Wagener et al (1999c). This may be due to differences in the method used; in contrast to our adaptive procedure, this reference value was determined by interpolating two measurements at fixed SNRs of  9 and  5 dB. Furthermore, the subjects in Wagener et al, 1999c experienced more training during the measurement. Other findings by Brand and Kollmeier (2002) in NH subjects and the same adaptive procedure as applied in the present study obtained results (SRT   6.3  0.52 dB SNR), supporting the linear estimate (1b). The linear relationship between PTA and SRT in noise was previously shown. Even early attempts by Fletcher and Steinberg (1930) imply a linear dependence of the SRT from hearing loss. More recently, Thiele et  al (2011) showed comparable results concerning the correlation of SRT and the PTA in the Göttingen sentence test with similar audiogram groups and PTA definitions as used in

\\

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

0

better performance Level [dB SPL]

40

10 5 0 better than predicted

–5 –10 –5

0

5

10 15 20 25 30 35 40 45 SRTpredicted [dB SNR]

//

> 45

Figure 6. SRTs predicted by SII of all groups A E. Upper panel: SRT predicted by SII as function of PTA (0.5–3 kHz). Lower panel: Measured SRTs with the OLSA at 65 dB SPL fixed noise level vs. the SRT predicted by SII. The dashed line depicts the equivalence of the prediction and the experimentally determined SRT.

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

78     N. Wardenga et al. the SII. A linear regression of group B provides a slightly better prediction (SD  1.10 dB). The most striking difference between SII and PTA predictions is that the SII does not distinguish between hearing losses in group A as the SII assumes that intelligibility is determined by the frequency specific maximum of the noise level given in dB HL and the hearing loss. The PTA, however, by definition also distinguishes between small degrees of hearing loss. Although it is obvious that audibility itself is indeed not determining the SRT in 65 dB noise, there are also possible reasons for assuming that there may be a correlation between the SRT in noise and low degrees of hearing loss. As described by Bjaradwj et al (2014) even a low degree of hearing loss might be an indicator for cochlear neuropathology. As a consequence supra-threshold processing deficits and a reduction of speech processing performance in noise can occur. Here the rather simple PTA concept may be helpful. This requires further investigation in future studies. Although we believe that in principle such ideas can be easily included in the SII, the PTA based approach shows further advantages. Like the SII, it provides a good estimate of the expected SRT in the entire range of hearing losses. The PTA, however, can be calculated very easily from the pure-tone audiogram, the interpretation is straightforward, and the method enables the identification of discrepancies between measurement results and predictions from the audiogram. It requires no additional software tools as the SII and the effort to use it in clinical routine is very low. Within its application range it may potentially serve in clinical practice to easily identify pathologies of more central origin, e.g. those affecting attention. Furthermore, the PTA can be used to check if the SRT was tested in an effective noise condition or rather in a quiet or hybrid condition. We believe that this practical aspect is important to encourage the use of SRT predictions in routine clinical practice by busy clinicians and dispensing audiologists. Testing subjects despite their hearing loss at a noise level of 65 dB is easily applicable in clinical trials and offers to test the communication abilities in a defined noise situation. In applications that explore the borders of the useful dynamic range, e.g. as in cochlear or in active middle-ear implants, the limitation to PTAs  47 dB HL creates a significant restriction. The commonly used method of testing the SRT at a noise level that was individually increased to medium loudness extends the usable range to more severe hearing losses. Here our results obtained using a fixed noise level can serve as a rule of thumb to estimate the range where the OLSA measures functionally the SRT in noise and where, for example, technical limitations are encountered.

Conclusion The OLSA can be applied to subjects with a wide range of hearing losses. With 65 dB SPL fixed noise presentation level the SRT is determined by listening in noise for PTAs  ∼47 dB HL, and above it is determined by listening in quiet.­­­­­­

Acknowledgements The authors would like to thank all the participants in this study, the students involved in the data collection: Sofie Katrina Stoll and Wojtek Stengert, as well as Birger Kollmeier for his input and support. This project was supported by the DFG Cluster of Excellence EXC 1077/1 ʽHearing4allʼ, the European regional development fund (EFRE), project ʽHurDigʼ and by the Lower Saxony Ministry of Science and Culture, project ʽAINʼ. Parts of the data had been presented at the 15th Annual Congress of German Audiological

Societies (DGA) in Erlangen, at the 16th Annual Congress of German Audiological Societies (DGA) in Rostock, and at the 84th Annual Meeting of the German Society of Oto-Rhino-Laryngology, Head and Neck Surgery in Nürnberg.

Note 1. ʽValidity determines whether the research truly measures that which it was intended to measure…ʼ (M. Joppe 2000, adapted from: Golafshani N. 2003, The Qualitative Report 8(4), 597–607). Declaration of interest: The authors report no conflicts of interest.

References Akeroyd M.A., Arlinger S., Bentler R.A., Boothroyd A., Dillier N. et  al. 2015. International Collegium of Rehabilitative Audiology (ICRA) recommendations for the construction of multilingual speech tests. Int J Audiol, 2015, Apr 29:1–6. [Epub ahead of print] ANSI S3.5-1997. Methods for Calculation of the Speech Intelligibility Index. New York: American National Standard Institute. Bharadwaj H.M., Verhulst S., Shaheen L., Liberman M.C. & Shinn-Cunningham B.G. 2014. Cochlear neuropathy and the coding of supra-threshold sound. Front Syst Neurosci, 8(26), 1–18. Brand T. & Kollmeier B. 2002. Efficient adaptive procedures for threshold and concurrent slope estimates for psychophysics and speech intelligibility tests. J Acoust Soc Am, 111, 2801–2810. Cameron S., Brown D., Keith R., Watson C. & Dillon H. 2009. Development of the North American listening in spatialized noise - sentence test (NA LiSN-S): Sentence equivalence, normative data, and test-retest reliability studies. J Am Acad Audiol, 20, 128–46. Fletcher H.M. & Steinberg J.C. 1930. Articulation testing methods. J Acoust Soc Am, 1(2.2), 17–64. Fletcher H. & Galt R.H. 1950. The perception of speech and its relation to telephony. J Acoust Soc Am, 22(2), 89–151. French N.R. & Steinberg J.C. 1947. Factors governing the intelligibility of speech sounds. J Acoust Soc Am, 19, 90. Glyde H., Cameron S., Dillon H., Hickson L. & Seeto M. 2013. The effects of hearing impairment and aging on spatial processing. Ear Hear, 34, 15–28. Gurgel R.K., Jackler R.K., Dobie R.A. & Popelka G.R. 2012. A new standardized format for reporting hearing outcome in clinical trials. Otolaryngol Head Neck Surg, 147, 803–807. Hagerman B. 1982. Sentences for testing speech intelligibility in noise. Scand Audiol, 11, 79–87. Hahlbrock K.-H. 1953. Über Sprachaudiometrie und neue Wörterteste (in German: Speech audiometry and new word-tests). Archiv für Ohren-, Nasen- und Kehlkopfheilkunde, 162, 394–431. Hey M., Hocke T., Hedderich J. & Müller-Deile J. 2014. Investigation of a matrix sentence test in noise: Reproducibility and discrimination function in cochlear implant patients. Int J Audiol, 53, 895–902. Hochmair-Desoyer I., Schulz E., Moser L. & Schmidt M. 1997. The HSM sentence test as a tool for evaluating the speech understanding in noise of cochlear implant users. Am J Otol, 18 (6), 83. Kießling J. 2000. Moderne Verfahren der Sprachaudiometrie (in German: Modern procedures in speech audiometry). Laryngol-Rhino-Otol, 79, 633–635. Kleine Punte A. & Van de Heyning P. 2013. Quality standards for minimal outcome measurements in adults and children. Cochlear Implants Int, 14, 39–42. Kollmeier B. & Wesselkamp M. 1997. Development and evaluation of a German sentence test for objective and subjective speech intelligibility assessment. J Acoust Soc Am, 102 (4), 2412–21. Kollmeier B., Lenarz T., Winkler A., Zokoll M.A., Sukowski H. et  al. 2011. Hörgeräteindikation und -überprüfung nach modernen Verfahren der Sprachaudiometrie im Deutschen (in German: Indication for and

Downloaded by [Purdue University Libraries] at 04:04 15 February 2016

verification of hearing-aid benefit using modern methods of speech audiometry in German). HNO, 59, 1012–21. Kollmeier B., Warzybok A., Hochmuth S., Zokoll M.A., Uslar V.N. et  al. 2015. The multilingual matrix test: Principles, applications, and comparison across languages: A review. Int J Audiol, (in press). Mueller H.G. & Killion M.C. 1990. An easy method for calculating the articulation index. The Hearing Journal, 43, 1–4. Müller-Deile J. 2009. Sprachverständlichkeitsuntersuchungen bei Kochlear­ implantatpatienten (in German: Speech intelligibility tests in cochlear implant patients). HNO, 57, 580–92. Thiele C., Sukowksi H., Lenarz T. & Lesinski-Schiedat A. 2011. Göttinger Satztest im Störgeräusch für verschiedene Gruppen von Schwerhörigkeit (in German: Göttingen sentence in noise for different audiogram classes). Laryngol-Rhino-Otol, 90, 1–7. Tschopp K. & Ingold L. 1992. Die Entwicklung einer deutschen Version des SPIN-Tests (Speech perception in noise). In: B. Kollmeier (ed.): Moderne Verfahren der Sprachaudiometrie. Heidelberg: Median-Verlag, 1992, pp. 311–329. Wagener K.C., Kühnel V. & Kollmeier B. 1999a. Entwicklung und Evaluation eines Satztests in deutscher Sprache. Part I: Design des Oldenburger Satztests (in German: Development and evaluation of a German sentence test). Z Audiol, 38, 4–15. Wagener K.C., Brand T. & Kollmeier B. 1999b. Entwicklung und Evaluation eines Satztests für die deutsche Sprache - Teil II: Optimierung des Oldenburger Satztests (in German: Development and evaluation of a

OLSA in unaided subjects with hearing loss      79 German sentence test - Part II: Optimization of the Oldenburg sentence test). Z Audiol, 38, 44–56. Wagener K.C., Brand T. & Kollmeier B. 1999c. Entwicklung und Evaluation eines Satztests für die deutsche Sprache - Teil III: Evaluation des Oldenburger Satztests (in German: Development and evaluation of a German sentence test. Part III: Evaluation of the Oldenburg sentence test). Z Audiol, 38, 86–95. Wagener K.C. 2004. Factors influencing sentence intelligibility in noise (dissertation) University of Oldenburg, BIS-Verlag, ISBN 3–8142– 0897–8. Wagener K.C. & Kollmeier B. 2004. Göttinger und Oldenburger Satztest. ZfA, 43 (3), 134–141. Wagener K.C. & Brand T. 2005. Sentence intelligibility in noise for listeners with normal hearing and hearing impairment: Influence of measurement procedure and masking parameters. Int J Audiol, 44 (3), 144–157. Wagener K.C., Brand T. & Kollmeier B. 2006. The role of silent intervals for sentence intelligibility in fluctuating noise in hearing: Impaired listeners. Int J Audiol, 45 (1), 26–33. Wittkop T., Zokoll M.A. & Kollmeier B. 2011. Bauartbedingte Unterschiede von Sprachverständlichkeitsschwellen bei Kopfhörermessungen in Ruhe (in German). 14th Congress of the German Society of Audiology, Germany, Jena: 10-12.03.2011 Zokoll M.A., Hochmuth S., Warzybok A., Wagener K.C., Buschermöhle M. et al. 2013. Speech-in-noise tests for multilingual hearing screening and diagnostics. AJA, 22, 175–178.