Abstract
For the audiological assessment of the speech-in-noise abilities of children with normal or impaired hearing, appropriate test materials are required. However, in Denmark, no standardized materials exist. The purpose of this study was to develop a Danish sentence corpus suitable for testing school-age children. Based on the 600 validated test sentences from the Danish DAT (
Children are often exposed to noise (e.g., in classrooms), which causes difficulties with speech recognition (e.g., Shield & Dockrell, 2003). In general, the adverse effects of noise on speech recognition diminish as children get older, but problems can be observed across childhood (Johnson, 2000; Soli & Sullivan, 1997; Werner & Boike, 2001). Consequently, reliable methods for assessing speech recognition in noise in children are essential, especially when deficits are suspected. In Germany, the ‘Oldenburger Kinder Satztest’ was developed for that purpose (Wagener et al., 2006). The Oldenburger Kinder Satztest consists of three-word pseudo-sentences that include a numeral, an adjective, and a noun (e.g. “four red flowers”). It has been found to be usable with children aged four and above (Neumann et al., 2012; Wagener et al., 2006). Another speech test that was designed for the use with children is ‘FreeHear,’ which is available in British English (Moore et al., 2019). In this test, three spoken digits are presented against a background of babble noise. Because of its simple sentence structure, this test has been found to be suitable for children as young as 4 years.
In Denmark, a number of speech materials are available for clinical and research purposes, for example, DANTALE I (Elberling et al., 1989) and DANTALE II (Wagener et al., 2003). DANTALE I includes lists of monosyllabic words for the measurement of discrimination scores. There are 4 lists containing 20 words each, which are considered suitable for children aged 5 and above. Furthermore, there is one list containing 20 words for younger children. However, none of these lists has been formally validated and standardized. DANTALE II includes lists of semantically unpredictable five-word sentences consisting of a name, a verb, a numeral, an adjective, and a noun each (e.g. “Kirsten bought four red flowers”). These lists were evaluated with normal-hearing adults, revealing significant training effects.
Another available Danish speech test is the conversational language understanding evaluation (CLUE) test (Nielsen & Dau, 2009). The sentence material used for the CLUE test stems from a large database, which includes everyday conversational sentences from Danish newspapers, magazines, books, and so on. The principles and procedure behind the CLUE test stem from the Hearing in Noise Test (HINT; Nilsson et al., 1994). In 2010, Nielsen and Dau (2011) developed a Danish HINT that was based on the speech material from the CLUE test. The Danish HINT consists of 10 test lists and 3 practice lists containing 20 sentences each. It has been shown to be a reliable measure of speech recognition for Danish adults with normal or impaired hearing (Nielsen & Dau, 2011). Although the reliability of the CLUE and Danish HINT has not been investigated with children, it is likely to be lower due to the properties of the speech material. First, the HINT material contains many words that are not part of the vocabulary of young children. Second, the sentences in the Danish HINT have different grammatical structures, some of which may be too complicated for children to understand. Third, the length of the sentences (five words) may exceed the memory capacity of younger children.
Another available set of Danish sentences is the DAT corpus, which is an open-set, low-context, multitalker speech corpus (Nielsen et al., 2014). The DAT corpus includes 600 unique sentences that have a fixed, simple structure. More specifically, they all make use of a carrier sentence that starts with a female name (i.e.,
To summarize, there are currently no standardized Danish materials that are suited to assessing speech recognition in noise in children. The purpose of this study was to address this shortcoming. In particular, the aim was to develop a set of test lists suited for 6- to 12-year olds. The aforementioned properties of the DAT corpus motivated us to use this material as the basis for the development of a ‘child-friendly’ DAT corpus called the børneDAT corpus. Ideally, this corpus should be characterized by small training effects, high test list equivalence, and low measurement uncertainty. In this study, we assessed these aspects by performing test–retest measurements with the created test lists over a time period of approximately 1 to 2 weeks.
Materials and Methods
Ethical approval for this study was obtained from the Regional Committees on Health Research Ethics for Southern Denmark.
Generation of Test Lists
For the compilation of the test lists, the 600 validated test sentences from the DAT corpus (Nielsen et al., 2014) were used. As pointed out above, all of these sentences have a fixed, simple structure. That is, they start with Dagmar (D), Asta (A), or Tine (T) and contain two short, neutral, and concrete keywords (nouns), for example, “Dagmar tænkte på
For this study, 220 sentences with keywords belonging to the vocabulary of a typical 6-year-old were selected. As part of the selection process, 2 audiological researchers and 1 researcher in the language development of children (i.e., three of the authors) individually went through all 600 sentences of the DAT corpus and assessed them in terms of their suitability for testing 6- to 12-year olds. Those sentences which all 3 researchers judged to be suitable were kept and combined into 11 lists containing 20 sentences each. In the original DAT study, the intelligibility of each sentence was assessed in a listening experiment involving 16 normal-hearing adult participants (Nielsen et al., 2014). The assessment was based on the assumption that sentence intelligibility can be quantified in terms of the signal-to-noise ratio (SNR) at which both keywords are correctly identified, and that more intelligible sentences are recognized at lower SNRs than less intelligible sentences. In the current study, the original sentence intelligibility assessments were assumed to be also valid for school-age children. Using the procedure from the original DAT study, sentences with relatively high and low intelligibility were counterbalanced at the beginning of each test list, while sentences with approximately equal intelligibility were used toward the end of each list (see Nielsen et al., 2014 for details). For a given list, only sentences uttered by the same talker (and thus starting with the same name) were used. In this manner, four D-lists (D1, D2, D3, and D4), three A-lists (A1, A2, and A3), and four T-lists (T1, T2, T3, and T4) were created. The 11 compiled test lists are provided in the Appendix.
Participants
A total of 20 typically developing, normal-hearing children (13 females) participated in the study. They were aged 6 to 12 years (mean: 8.7 years). Their parents provided written informed consent, and the children received a gift card at the end of the study.
All participants fulfilled the following inclusion criteria: (a) normal middle-ear function, (b) pure-tone hearing thresholds ≤ 25 dB HL at all standard audiometric frequencies from 125 to 8000 Hz, (c) normal speech discrimination in quiet, (d) native Danish speakers, (e) normal language development, and (f) normal cognitive function. Otoscopy and tympanometry were performed to examine the outer and middle ears of all participants. Children with any type of obstruction or infection in the ear canal and/or type-B or type-C tympanograms were excluded. Standard pure-tone audiometry was carried out using supra-aural headphones. Next, speech discrimination in quiet at the most comfortable level was tested using the DANTALE I material (Elberling et al., 1989). Listeners with discrimination scores <90%-correct were also excluded. Language development of the children was assessed using the Peabody Picture Vocabulary Test (Dunn et al., 1965), which confirmed normal language skills for all participants. Cognitive development was assessed based on parental reports. In addition, a custom-made questionnaire was administered that included questions related to the child’s mother tongue, whether the child was monolingual, and the income and education level of the child’s parents. All participants in this study were monolingual, native Danish speakers and came from families with middle to high incomes.
Apparatus and Procedures
All measurements were conducted in a soundproof booth. To evaluate the 11 created test lists in terms of their perceptual similarity and reliability, SRT measurements were made. The speech stimuli were presented diotically in stationary speech-shaped noise through supra-aural, free-field equalized headphones (Sennheiser HDA200). The speech-shaped noise was talker-specific. The order of the test lists was balanced across the participants. The starting level of the speech signal was 67 dB SPL. The level of the noise was fixed at 60 dB SPL. The equipment was calibrated using a 01 dB FUSION sound level meter and a GRAS 43AA-S2 CCP ear simulator kit. The SRTs were measured using the adaptive procedure from the Danish HINT (Nielsen & Dau, 2011). The participants were verbally instructed to repeat the two keywords in each sentence. In case of any doubts, they were encouraged to guess. Responses were scored as correct if both keywords were repeated accurately.
Before the start of the actual measurements, all participants performed one SRT measurement in quiet and two SRT measurements in noise. The lists used for these purposes were training lists from the original DAT material. A short break was included after the first five SRT measurements and whenever a participant felt tired. A set of retest measurements was made on average 10 days (range: 5–19 days) after the first set of measurements.
Statistical Analyses
The collected data were analyzed using SPSS (IBM) version 25. To begin with, the consistency of each child’s test–retest data was assessed based on squared difference scores and scatter plots. This resulted in the data of one child being excluded from all subsequent analyses because of clear inconsistencies. To verify the equality of variance in the datasets, Levene’s test was used. This showed equality of variance in the datasets for all test lists. Shapiro–Wilk’s test, normal
Results
Perceptual Validation of the Generated Test Lists
Figure 1 shows the mean list SRTs for the two visits. The grand average SRT across all lists and participants was −2.2 dB SNR with an across-subject standard deviation (
Mean List SRTs for the First (Black) and Second (Gray) Visit. Error bars show ±1
Given that the sentences of the D-, A-, and T-lists were uttered by three talkers with different voice characteristics, the SRT measurements were analyzed in terms of a potential talker effect. The overall mean SRTs of the D-, A-, and T-lists were found to be −1.5 dB SNR, −2.8 dB SNR, and −2.3 dB SNR, respectively. A one-way ANOVA comparing the mean SRTs of the D-, A-, and T-lists showed a significant effect,
Results of Post Hoc Tests Comparing the Mean SRTs of the D-, A-, and T-Lists.
To investigate the perceptual similarity of the A- and T-lists, a two-way repeated-measures ANOVA with the within-subject factors visit and list was carried out. This showed statistically significant effects of list,
To investigate the perceptual similarity of the four D-lists and the T1-list, another two-way repeated-measures ANOVA was carried out. This showed statistically significant effects of list,
Definition of Training and Test Lists
Based on the statistical results above, two sets of lists were defined: one for training (D1, D3, D4, and T1) and one for testing (A1, A2, A3, T2, T3, and T4). Figure 2 shows the mean SRTs of the lists for the two sets. For the six test lists, the grand average SRT was −2.6 dB SNR, the average test–retest improvement 0.6 dB, the within-subject
Mean SRTs for the Two Sets of Perceptually Equivalent Test Lists (Test Lists: Black Squares; Training Lists: Gray Diamonds). Error bars show ±1
Age Effect
Given that the study participants covered a relatively wide age span (6–12 years), the effect of age on the SRT results was also tested. Figure 3 shows a scatter plot of age against the grand average SRT (calculated across all 11 test lists). As expected, older children achieved lower (better) SRTs compared with younger children. The relationship between age and mean SRT was statistically significant,
Scatter Plot of Age Versus Grand Average SRT With a Least-Squares Regression Line. SRT = speech recognition threshold; SNR = signal-to-noise ratio.
Discussion
The current study aimed to develop a Danish speech material, which is suitable for assessing speech recognition in noise in school-age children. More specifically, the objective was to develop a set of test lists with small within-subject and between-list variation that can be used for performing SRT measurements with 6- to 12-year olds. Eleven lists comprising 20 sentences each were compiled based on the validated sentence material from the DAT corpus (Nielsen et al., 2014). Test list equivalence and measurement reliability were examined with the help of 20 typically developing, normal-hearing Danish children.
To assess the properties of the børneDAT material, the obtained results were compared with those for the Danish HINT (Nielsen & Dau, 2011) and DAT material (Nielsen et al., 2014). In this study, the grand average SRT of the 11 developed lists was −2.2 dB SNR, which is comparable to the mean SRT of the Danish HINT obtained with 16 normal-hearing adults (–2.5 dB SNR; Nielsen & Dau, 2011). The within-subject
Since the sentences of the D-, A-, and T-lists were uttered by three different talkers, we considered the influence of talker on the results. We found that the D-lists resulted in significantly higher mean SRTs than the A- and T-lists. As part of the development of the original DAT corpus, the effect of talker was also considered. For SRTs measured with adult listeners and a speech-on-speech masking paradigm, Nielsen et al. also observed higher SRTs for talker D. Overall, these findings suggest that the D-sentences are slightly less intelligible than the A- and T-sentences. Nielsen et al. (2014) did not investigate the cause of this, but a likely explanation is differences in pronunciation between the talkers.
Ideally, the test lists of a given speech corpus should result in very similar SRT measurements, so they can be used interchangeably. Based on our results, the developed material includes six lists (A1, A2, A3, T2, T3, and T4) with equivalent mean SRTs that we propose to use for test purposes. The mean test–retest improvement for these lists was 0.6 dB, corresponding to that observed for the Danish HINT. The within-subject
Regarding the effects of age, this study confirmed that younger children struggle more to understand speech in noise compared with older ones. This is consistent with a large body of research comparing the effects of noise on younger versus older children (e.g., Johnson, 2000; Soli & Sullivan, 1997; Walker et al., 2019; Werner & Boike, 2001), and can be traced back to developmental changes in terms of language skills of school-age children (e.g., Firmansyah, 2018). At a more general level, this finding implies that since younger children understand speech especially poorly in noisy environments it is particularly important to provide classrooms with good acoustical conditions for them.
The developed speech material is publicly available and can be used in speech-based research with Danish 6- to 12-year olds. Nevertheless, some limitations should also be noted. First, to allow the developed speech material to be used in clinical practice, age-specific normative data need to be collected. Second, the validation we performed was restricted to normal-hearing children. Future work will therefore also examine the properties of the speech material with children with hearing loss.
Conclusions
Based on the validated sentence material from the Danish DAT corpus, a new set of sentences aimed at Danish 6- to 12-year olds was constructed and evaluated in terms of its reliability. Six test lists were found to be equivalent in terms of their mean SRTs. A small training effect was observed, suggesting that the lists can be reused after 1 to 2 weeks. Four lists considered to be suitable for training purposes resulted in mean SRTs that were on average 1 dB higher than the mean SRTs of the other six lists, but otherwise equivalent and usable. Overall, the developed material seems therefore suitable for research studies with Danish 6- to 12-year olds.
Appendix
Developed Test and Training Lists
Test Lists
1. en kaptajn, en saks; 2. en pingvin, en streg; 3. en træsko, et skjold; 3. en fisk, en knæskal; 4.en skovtrold, en mark; 5. en biograf, en fælde; 6. et arbejde, en paryk; 7. en plade, en byfest; 8. en weekend, en gaffel; 9. en hytte, et papskilt; 10. en zebra, en børste; 11. en boble, et græskar; 11. en ost, et postkort; 12. en vandpyt, en tønde; 13. et vindue, en jagthund; 14. en bogreol, en konge; 15. en port, en paraply; 16. en vask, en flodhest; 17. en pelikan, en suppe; 18. en isbjørn, en kvinde.
1. en ketsjer, en tånegl; 2. en reklame, en kok; 3. en diamant, en cykel; 4. et lotteri, en lygte; 5. en skovsø, en dåse; 6. en klub, et bolsje; 7. en svamp, en kagedej; 8. en solsort, en potte; 9. en bjørn, en fodbold; 10. et eventyr, en duks; 11. et batteri, en klods; 12. en opgave, et spark; 13. et hjerte, en prins; 14. en hårtot, en støvle; 15. et paprør, en æske; 16. et smykke, en elefant; 17. en trøje, en bageovn; 18. et tebrev, et telt; 19. en kjole, et blåbær; 20. en due, et posthus.
1. en brand, et værksted; 2. en færge, en gulvklud; 3. et kamera, en stork; 4. en busk, en undulat; 5. en tyr, en kantsten; 6. en garage, en tærte; 7. en tavle, en vaffel; 8. en salat, en skramme; 9. en cirkel, et hylster; 10. et blad, et korthus; 11. en guldfisk, en kost; 12. en terning, en skøjte; 13. en fletning, en cigar; 14. en form, et dørskilt; 15. en småkage, en rejse; 16. et stempel, en smil; 17. et dyr, en kridtstreg; 18. en vinter, en storby; 19. en pige, en trompet; 20. en vikar, en sandal.
1. et træhus, en sovs; 2. et æble, en krabbe; 3. et styr, en grøntsag; 4. en landsby, en balje; 5. en tepose, en hest; 6. et postbud, en kant; 7. en gang, et fjernsyn; 8. en tur, en limstift; 9. et æsel, en tromme; 10. et krybdyr, en pande; 11. en dessert, en fløjte; 12. en flagstang, en hvalp; 13. et skakspil, en abe; 14. en skammel, en dagbog; 15. en ble, en bjergtop; 16. en ølkasse, en giraf; 17. et kødben, en stemme; 18. et fortov, en frakke; 19. en drøm, en sløjfe; 20. en regnskov, en heks.
1. en finger, en skovtur; 2. et hus et, vinglas; 3. et spring, en fortand; 4. et kor, et fodspor; 5. et træ, en krystal; 6. en kanon, en stolpe; 7. et tog, et spædbarn; 8. en sportsmand, et bord; 9. en ballon, et stopur; 10. en præmie, en klasse; 11. en skorsten, et føl; 12. en påfugl, en buket; 13. en sværd, en tulipan; 14. en kran, en delfin; 15. et palads, en dyrlæge; 16. et pindsvin, et flag; 17. en agurk, et vandløb; 18. en bagdør, en fridag; 19. et håndtryk, en ørken; 20. en kommode, en frugt.
1. en sportsvogn, en sild; 2. en uniform, et bælte; 3. et apotek, et sving; 4. et vejskilt, en tiger; 5. en fe, en jordklump; 6. et drivhus, en fugl; 7. et juletræ, en tand; 8. en tekop, et fnis; 9. en hamster, en prøve; 10. en grøft, en vulkan; 11. en hveps, en fyrste; 12. en kålorm, en ferie; 13. en lasso, et bakspejl; 14. et apparat, en dyne; 15. en snog, et græsfrø; 16. en dans, en lastbil; 17. en parasol, en stige; 18. en ske, et stålrør; 19. en tråd, et skumbad; 20. en skov, en tegning.
Training Lists
1. en teske, en næse; 2. en madkurv, en kogle; 3. et legehus, en rift; 4. en gorilla, en datter; 5. en sytråd, en strand; 6. en pels, en kontakt; 7. en klovn, en høstak; 8. en trappe, en tilbud; 9. en smørklat, en vest; 10. en bilkø, en vifte; 11. en fad, en blinklys; 12. en gård, et græsstrå; 13. en troldmand, en kasse; 14. en torsk, en bande; 15. et soltag, en sauna; 16. en bro, et fængsel; 17. et net, en kronprins; 18. en plads, et kapløb; 19. en pære, et kortspil; 20. en film, en knivspids.
1. en sprøjte, et job; 2. en brandmand, en flue; 3. en nabo, en maskine; 4. en vejkant, en mose; 5. en kiks, en jungle; 6. et knaphul, en krog; 7. en skorpe, en rygsæk; 8. en malkeko, et svar; 9. en skat, et kæledyr; 10. en glasskål, en sut; 11. en sok, en sporhund; 12. en sommer, en fedtplet; 13. en snabel, et skilt; 14. en elastik, en gnist; 15. en tabel, et spil; 16. en fest, et bilhorn; 17. en pejs, et snobrød; 18. en grotte, et betræk; 19. en økse, en rundkreds; 20. en ispind, en stribe.
1. et rensdyr, et slot; 2. en sæk, et sejlskib; 3. en frisør, en klokke; 4. et bål, en kænguru; 5. et hindbær, en mus; 6. en festsal, en rist; 7. en krølle, en strømpe; 8. en kamel, en hilsen; 9. en gråspurv, en seng; 10. en båd, et springvand; 11. en mursten, en kappe; 12. en gårdsplads, en mønt; 13. en lakrids, en doktor; 14. en ankel, en rovfugl; 15. en tryne, en blyant; 16. en ridetur, en billet; 17. en sko, en popsang; 18. en adresse, et krus; 19. en formand, en synål; 20. en nisse, en trykknap.
1. en flaske, en rosin; 2. en klud, en statue; 3. en gåtur, en drage; 4. et forår, et punktum; 5. et halsbånd, en spids; 6. et album, en pose; 7. en gulerod, et skab; 8. en skurvogn, en hætte; 9. en skjorte, et skib; 10. en fabrik, en figur; 11. et kort, en trætop; 12. en salami, en sofa; 13. en række, en tunfisk; 14. en sportshal, et dæk; 15. en abrikos, en sti; 16. en skovsnegl, en bus; 17. en fjer, en kornmark; 18. en robåd, en skinke; 19. et ønske, et lyntog; 20. en tante, en sørøver.
Footnotes
Acknowledgments
This research was supported by a PhD stipend from the William Demant Foundation. The authors wish to thank Signe Hjorth Fogh and Nicoline Gotholt Madsen for their help with the data collection. The børneDAT material is available upon request to the corresponding author.
Declaration of Conflicting Interest
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
