Efficient and sensitive hearing assessment based on speech perpeception

Besides the gold standard pure-tone audiometry, it is highly relevant to assess a person s supra-threshold speech perception skills, especially in background noise, as it approaches our functional hearing ability in a daily communicative environment. In case of a sensorineural hearing loss, an impairment leading to both the attenuation and distortion of incoming sounds, speech perception problems are highly likely to arise. A relationship thus exists between the speech reception threshold (SRT) in noise and the pure-tone thresholds (PTT), although not perfect. The main influencing factors are (1) individual differences in (peripheral or central) supra-threshold versus threshold auditory processing, (2) cognitive and linguistic skills, (3) the redundancy in speech samples, and (4) the measurement error on both behavioral measures. In this PhD project, it was investigated how different types of speech and noise material may influence the SRT-PTT relationship and may lead to new possibilities, new applications of a speech-in-noise test for efficient and sensitive hearing assessment.In the first part of this PhD thesis, including the first three studies, the efficiency and sensitivity for screening purposes was investigated. Both for detecting age-related hearing loss, which is slowly progressing and often undiagnosed, and for noise-induced hearing loss, where periodic screening in high-risk populations may prevent further damage, there is a need for a robust, simple, fast, and sensitive hearing screening technique. When conducting a supra-threshold speech-in-noise test, neither a sound-proof booth nor a high-quality, well-calibrated audiometer are requisite, and it can be implemented as a quick, automated self-test. This makes a speech-in-noise test highly suitable for screening purposes. Smits et al (2004) were the first to develop their Dutch Digit Triplet speech-in-noise self-screening test through home telephone. In the first study, a Digit Triplet test was developed in French, for both telephone and broadband headphone presentation. Due to the separate optimization of the speech material based on the intelligibility of the individual digits, a steep psychometric curve (17 and 27%/dB, respectively) and a low measurement error on the SRT (0.7 and 0.6 dB, respectively) were achieved for both versions. Furthermore, the telephone version was validated in hearing-impaired listeners and yielded a strong correlation of 0.77 with the pure-tone average at 0.5, 1, 2, and 4 kHz (PTA0.5,1,2,4). With 20 000 calls within one month after the nationwide release in France, the test also proves to fulfill the need for an easily accessible hearing screening test. In the second study, the sensitivity of the broadband Digit Triplet test for detecting (early-stage) high-frequency hearing loss was investigated in a group of 118 noise-exposed workers. The 84 Dutch-speaking participants showed a very strong linear relation between the SRT and PTA2,3,4,6 (R=0.86). The sensitivity and specificity to detect mild high-frequency hearing loss were 92% and 89%, respectively. The results of the 34 French-speaking participants showed a highly similar trend. Third, it was explored if a further improvement could be achieved by changing the speech items of the test (9 consonant-vowel-consonant (CVC-) words all with the same vowel) and/or by using a low-pass (LP) filtered version of the standard speech-shaped noise. Within the same group of noise-exposed listeners, the CVC test showed a higher measurement error and a lower between-subject variation than the Digit Triplet test, leading to a lower sensitivity. The CVC test in LP-noise, on the other hand, yielded a sensitivity improvement due to the large increase in between-subject spread. However, there was no improvement compared with the original Digit Triplet test. In the second part, three types of speech-in-noise tests, all developed with a different purpose, were compared to each other: a common everyday sentence test, a closed-set sentence test (Matrix), and the Digit Triplet test. Forty-nine normal-hearing and 69 mildly-to-moderately hearing-impaired listeners, recruited from four French-speaking regions in Belgium and France, participated. No regional effects on any of the three types were found. Mutual correlations between the three tests were high (R>0.81). The Digit Triplet and Matrix test, both with a low contextual redundancy, showed the highest discriminative power, especially between the normal-hearing and mildly hearing-impaired listeners.nrpages: 188status: publishe