Interpretation of the norms of HINT with a female speaker and comparison with the data by HINT with a male speaker and OLSA
According to Brand and Kollmeier, standard deviation values of 1 dB SNR are important to achieve high validity of Speech-in-Noise-Tests [
6]. The standard deviations of the norms of the German HINT with a female speaker are in the mentioned range about 0.54 dB (NCS), 1.02 dB (NF), 1.15 dB (NR) and 1.55 dB SNR (NL) and comparable with the SD for the German HINT with a male speaker (NF 0.8 dB, NR 0.9 dB, NL 0.7 dB and NCS 0.5 dB SNR). The average standard deviation in Quiet are lower but still comparable between both speakers (female 2.4 dB, male 2.6 dB). Expect the condition noise right, there was no significant difference between the norms by the female speaker and the norms by Joiko et al. for the male speaker. The speech material including sentence-list-classification of both tests and the studies included subjects are nearly equal. The subjects were young (mean age 26.8 years, Joiko et al. 26.7 years), mostly students and normal hearing listeners (pure-tone hearing threshold levels around 6 dB) [
17]. Similar to the results found by Yoho et al. the homogeneity of the subjects seems to be responsible for the mentioned non-significant effects between the female and male HINT and, furthermore, for the non-significant effect of the subjects’ gender on the speech intelligibility score [
37]. Unnoticed interindividual subject difference could be responsible for the significant difference between the SRT results of the conditions NR by the female speaker and NR, NL and the NCS by the male speaker of this study compared to the norms from Joiko et al. (NR: − 14.19 (female norms) and − 14.13 (male speakers results) versus − 13.60 dB SNR (male norms), NL: − 14.73 versus − 13.70 dB SNR and NCS: − 10.13 versus − 9.88 dB SNR) [
17]. Furthermore, Joiko et al. calculated the male HINT norms by 40–91 SRT results contrary to the 24 SRT data for determination of the norms of the female HINT [
17].
Nevertheless, the threshold norms for the female speaker are generally comparable to the norms for the male speaker. Using a second order polynominal trendline, the slope of the Performance intensity function by female HINT is about 8.84% per dB and by male HINT about 9.8% per dB. The reduced slope for the female speaker could be the consequence of the ‘fixed final HINT sentence from selected 334 natural sentences. Using the same 240 sentences for female HINT was more important than eliminating the ‘poorest’ and ‘best’ intelligible sentence resulting by the slope of the PI function. No adjustments on the selected sentences by female sentences material could be made. The following adjustment of the RMS level of the final female HINT sentences was done based on the results for the female PI measurements.
Both PI slopes are lower than the average slope of the thirteen other languages by Soli and Wong about 10.3% per dB but within the range of the first SD around the mean [
30]. The German PI slopes were calculated as a part of the measurements to evaluate the difficulty of the sentences. Furthermore, the German slopes were calculated based on word scoring, whereas the slopes of the other languages were calculated based on sentence scoring protocols. Nevertheless, the PI function German female and male HINT are comparable to the slope of the mentioned thirteen other languages. The slopes are still lower than OLSA’s introductory slope of about 17.1% per dB. Larger changes in speech sound pressure or SNR are required by HINT compared to OLSA to achieve differences in speech intelligibility [
33].
Harianawala et al. compared the slope of the PI function of the American matrix test and the American HINT determined by fixed SNRs including ten subjects with hearing aids. Contrary to the slopes by normal hearing subjects, the slope of the HINT was steeper (∼ 14% per dB) than the slope of the matrix test (11% per dB). The HINT seems to have small advantages by detecting differences in speech intelligibility resulting from variations in hearing aid processing [
14]. Further studies for detailed results of the PI slope for the German HINT including normal hearing subjects as well as subjects with hearing aids are recommended.
In their study, Soli and Wong compared the SRT means and standard deviations of thirteen languages [
30]. The average SRT for other languages were − 3.9 dB SNR (NF), − 11.2 dB SNR (NR), − 11.3 dB SNR (NL) and − 7.6 dB SNR (NCS) (refer to Table
1, Soli and Wong 2008) [
30]. The noise condition norms of the German HINT with a female speaker (shown in Table
3) are significantly lower (better) than the average thresholds for other languages [
17,
30]. Hochmuth et al. investigated the speech recognition of matrix sentences spoken by German/Russian and German/Spanish bilingual speakers [
16]. The authors found both language and speaker specific effects for the German/Spanish bilingual speakers. German SRTs were generally 2–4 dB lower than Spanish SRTs. Specific articulation and grammatical features and the structure of the Spanish language could responsible for these effects [
16]. Additional studies using bilingual HINT speakers are useful to evaluate the SRT difference made between German and other languages and to examine possible specific features and characteristics of the German language. Nevertheless, speaker-specific characteristics affected speech intelligibility in noise more than language-specific characteristics [
16,
30]. Especially conditions in Noise seems to be relevant. The mean values of the conditions in Quiet of German female and male HINT are in the range of the mean SRTs between 15.3 dB and 25.9 dB (mean 18.4 ± 3.5 dB of the other languages) across other languages [
30].
A significant interaction between the SRT values of the conditions NF and NCS were expected, because the score is calculated by threshold results of the investigated noise conditions. The significant correlation between the threshold values of the conditions NF by the female and male speaker supports the observation about the non-significant influence of speaker’s gender on speech intelligibility. Instead, the speech material seems to have a relevant impact on speech intelligibility scores showing non-significance correlation between the SRT results of HINT and OLSA. The results by OLSA must be evaluated talking into account of the test method and setting used. The results of the OLSA (mean − 5.80 dB SNR) differ significantly from the norms by Wagener et al. about − 7.1 dB SNR. The results of the present study were evaluated under headphone conditions and after a training session of 30 sentences compared to the data by Wagener et al. who recommended and used 6 list per 20 sentences for training and sound field conditions [
33]. Results in sound field conditions are influenced by the individual acoustics effects of the audiology test environment [
30]. The aim of the study was to develop norms for the German HINT with female speakers and to compare these norms with the German HINT with male speakers under headphone presentation. To be comparable and to eliminate variable effects of room acoustics, a headphone setting was also used by OLSA. Nevertheless, the completion of the training procedure of the OLSA by one list compared to the mentioned six lists by Wagener could be responsible for the significant difference between the resulting OLSA mean values compared to the norms by Wagener et al.. The HINT measurements were performed by completing the training session using one 20-sentence-list. Similar to the results by the Swedish HINT by Hällgren et al., a short training session seems to be sufficient for normal hearing subjects to get familiar with the test procedure [
13]. The measurements of each subject included in total 12 lists per 20 sentences. A high level of concentration was required of the subjects during the study which lasted about 2 h. As a consequence, shorter training sessions were used. Nevertheless, the comparability of the resulting SRT values of OLSA and HINT must be considered in the context of their individual test environment, the difference within the test procedure and the scope of training.
The most important variable of speech-in-noise tests concerns the speech material (e.g., numbers, words, rhyme and sentence) and in the present study focuses specifically on sentence structure and vocabulary. The limited vocabulary (50 words) of the OLSA being referred to as a ‘semi-open’ test. Repeated words could be recognized and are easier to guess. Mentioned influence could distort the results like moving the SRT to lower (better) SRT values. The phoneme distribution of the OLSA base list represents the phoneme distribution of the German language. The OLSA Test was developed in 1999 [
33‐
35]. It’s reference data for the phoneme distribution of the utilized words and names are from the 1970s are outdated [
34]. Especially some used names are outdated and, therefore, easy to remember for younger subjects. The phoneme distributions of 2019 generated HINT sentence within each of the twelve 20-sentence-lists did not differ significantly [
17]. Further studies are planned to investigate the ICRA recommendations about language and contextual skills and adult specific phoneme distribution of the German language [
2].
The vocabulary size of the HINT words is more extensive compared to the OLSA. Nevertheless, matching with childlex database limits the ability of speech materials using vocabulary of 6–8-year-old children [
17]. Further studies are needed to verify the German HINT with a female speaker within younger subjects as well as normal hearing adults to evaluate a test–retest reliability measurement.
In addition, cognitive effects and the working memory capacity influences speech intelligibility. Rudner et al. have shown that these effects are larger by Matrixtest than everyday sentences test like the HINT [
27]. Matrixtest are semantically unpredictable and less redundant. Constrained structures and the content of Matrixtests make guessing more difficult compared to meaningful and simple HINT sentences. With age, the linguistic and context-related skills increase, while working memory capacity decreases [
27]. In clinical routine, speech-in-noise-tests are used to evaluate changes in individual speech recognition that may lead to an indication for hearing aids. Mean age of clinical tested subjects is, therefore, mostly older than 26.8 years. Uslar et al. evaluate within their study the influence of three different types of linguistic complex sentences on speech reception in noise for younger and older subjects. Even if their study shows less relevance of linguistic complexity on speech reception in noise, the authors are sure about the influence of linguistic complexity as well as the age and the hearing conditions of the subjects. Further comparative HINT and OLSA studies should include different groups of subjects (e.g., younger and older subjects) to assess the influence of subject age and the linguistic complexity across the used sentences on speech recognition [
32]. The comparison of the results of the first and the second measurements of OLSA and HINT showed larger training effects for the OLSA (absolute difference between the mean SRT of the first and second measurement: OLSA: 0.72 dB SNR, female HINT: 0.61 dB SNR, male HINT: 0.21 dB SNR). The HINT seems advantageous to assess speech recognition more quickly and without extensive training sessions, but in context of its limited number of lists and possible learning procedures with repeated use.
Following HINT studies should be done to investigate the effect of repeated lists across test sessions on the speech intelligibility score. After completion of the training procedure, the OLSA seems suitable for frequent measurements in the hearing aid fitting process because of its random repeatability. Supporting the first notices by Hällgren et al., the test environment, the test methods including speech material and training effects and the included subjects’ have an influence on the speech recognition in noise [
13].
Effect of speakers’ gender on speech intelligibility score
Further aim of the present study was to investigate the effect of female and male speakers voice on speech recognition threshold of normal hearing subjects. A lot of studies investigated the influence of speakers’ gender and its acoustic phonetic characteristics on speech intelligibility. Hazan and Markham (2004) [
15] and Bond and Moore (1994) [
4] showed that longer word duration and slower speaking rates are correlating with better speech intelligibility scores. There was no correlation with the mean F0 range in the study by Hazan and Markham [
15]. The mean speaking rate of both male speakers in the present study is significantly lower (233 syllable/ min, OLSA and 251 syllable/ min, male HINT) compared to female speakers who speak about 287 syllable/min [
34]. Nevertheless, the reached SRT values did not differ between the female and male HINT speakers. Similar results were found by Bradlow et al. (1996) by investigating sentence intelligibility in quiet. A slower speaking rate and the mean F0 did not correlate with better speech intelligibility scores [
5]. Hazan and Braida showed that speakers’ timing, the precision of articulation and the long term spectra in the 1–3 kHz region correlate significantly with better intelligibility of female speakers [
15]. Krause and Braida emphasize the importance of a clear voice compared to a slower speaking rate. By subjective judgment, the female HINT voice seems to be more clear and phonetical accentuated than the male voice. Nevertheless, the SRT norms do not differ. The influence of the “intrinsically clear” speaking style on speech recognition scores seems to be less important in the present study compared to the data by Krause and Braida [
22]. Further studies should investigate the mentioned subjective findings about HINT speakers speaking style.
Bradlow et al. (1996) found that female speakers were significantly more intelligible than male speakers using sentences in quiet conditions. The authors explained their data by higher F0 range and wider dispersion of the phonetic vowel space of female voices compared to male voices [
5]. The results by Bond and Moore and Hazan and Markham supports the relevance of a wider vowel space for better speech intelligible scores [
4,
15]. The mean F0 range and the dispersion of vowel space (F1 and F2 values of the vowels /a:/ and /i:/, F2 value of /u:/ and the difference between F2 and F1of /i:/) differ significantly between female and male voices by wider range of female speaker. The vowel size as well as the F1 and F2 range are larger at female speakers’ voices compared to male speaker’s voices (refer to Table
2). Nevertheless, both speakers’ intelligibility scores did not differ significantly. The mentioned acoustic parameters seem to have no relevant influence on the SRT results in the present study. Still, the statement must be interpreted in the background about the limited numbers of analyzed vowels (
n = 3) and sentence (
n = 36) and the linguistic complexity (refer to Uslar et al.) based on phoneme.
In 2013 Ahrlich developed the German female Matrix test. The mean SRT is about – 9.3 dB SNR compared to the normative SRT by Wagener et al. which is about – 7.1 dB SNR. Despite an extensive analysis of the speech-acoustic parameters by six different speakers (female OLSA; male OLSA and four bilingual speakers—two female, two male whose speech material and SRT values originate from Hochmuth et al. 2013), the authors could not determine a clear connection between the resulting mean SRT and discussed acoustic parameters (mean F0, mean F0 range, vocal range by /a:/, /i:/ and /u:/, vowel triangle, F1-/ F2-range, speaking rate). Similar to this study, the authors could not present any correlation between the speakers’ gender and the resulting SRT values [
1].
In conclusion, the present study supports the illustrated inconsistency of the mentioned studies. There are no specific acoustic–phonetic characteristics that force higher intelligibility scores. Furthermore, especially the presentation modus—namely, within and without noise, has an important influence on speech intelligibly score and speakers intelligibility as well. Different speakers use different combinations and strategies of the mentioned acoustic–phonetic parameters to achieve high intelligibility scores. High interindividual variabilities in the speaker’s speaking style obtain the difficulty that a single speaker may never be fully representative of his or her current gender as a whole The female voice of the German HINT had a rather lower mean F0 about 179 Hz compared to literatures mean F0 about 200 till 220 Hz [
31]. Despite a large difference between the mean F0 of the female and male HINT speakers’ voice (105 Hz, male and 179 Hz female voice) the mentioned lower mean F0 of the female voice offers a possible explanation for the equality and non-significance difference between both HINT norms of three of the four tested conditions (NF, NL and Quiet) and the calculated NCS.
Calculating speech intelligibility scores by a single speaker while knowing about the diversity of voices and their differences in acoustic phonetic characteristics in the everyday acoustic environment seems to be difficult. Instead, speech intelligibility should be reviewed by comparing results of different speakers of different genders. Kelly et al. developed in 2017 a mixed gender, multi-speaker matrix sentence test in Australian English. Ten speakers, five women and five men, some of them professional actors and some amateur actors, were part of the recording process. The norms and slope of the multi-speaker matrix sentence test were comparable to reference values obtained from other matrix test involving single speakers (e.g., German and Swedish matrix test [
18]). Speech recognition scores investigated by speech-in-noise-tests within both speakers are more realistic presentations of the everyday acoustic environment, although there was no significant reference of the HINT talker gender for the norms, except of one condition.
Finally, it is recommended and necessary to integrate further speech-in-noise-tests, especially tests with female speakers in clinical audiological assessment. Referring to Harianawala et al. (2019), the clinical decision for the use of one of the multitudes of mentioned speech-in-noise-tests will depend on different variables and factors. Test-related factors are about the indication and aim of the test, about the including subjects and their linguistic and cognitive skills, the availability of the speech material and the test time, about the experience within the recommended test and the expected or known power of speech intelligibilities disorder [
14].