Towards implementation of adult hearing screening tests that can be delivered via a mobile app, we have recently designed a novel speech-in-noise test based on the following requirements: user-operated, fast, reliable, accurate, viable for use by listeners of unknown native language and viable for testing at a distance. This study addresses specific models to (i) investigate the ability of the test to identify ears with mild hearing loss using machine learning; and (ii) address the range of the output levels generated using different transducers. Our results demonstrate that the test classification performance using decision tree models is in line with the performance of validated, language-dependent speech-in-noise tests. We observed, on average, 0.75 accuracy, 0.64 sensitivity and 0.81 specificity. Regarding the analysis of output levels, we demonstrated substantial variability of transducers’ characteristics and dynamic range, with headphones yielding higher output levels compared to earphones. These findings confirm the importance of a self-adjusted volume option. These results also suggest that earphones may not be suitable for test execution as the output levels may be relatively low, particularly for subjects with hearing loss or for those who skip the volume adjustment step. Further research is needed to fully address test performance, e.g. testing a larger sample of subjects, addressing different classification approaches, and characterizing test reliability in varying conditions using different devices and transducers.

Development and Evaluation of a Novel Method for Adult Hearing Screening: Towards a Dedicated Smartphone App

Polo E. M.;Zanet M.;Barbieri R.;Paglialonga A.
2021-01-01

Abstract

Towards implementation of adult hearing screening tests that can be delivered via a mobile app, we have recently designed a novel speech-in-noise test based on the following requirements: user-operated, fast, reliable, accurate, viable for use by listeners of unknown native language and viable for testing at a distance. This study addresses specific models to (i) investigate the ability of the test to identify ears with mild hearing loss using machine learning; and (ii) address the range of the output levels generated using different transducers. Our results demonstrate that the test classification performance using decision tree models is in line with the performance of validated, language-dependent speech-in-noise tests. We observed, on average, 0.75 accuracy, 0.64 sensitivity and 0.81 specificity. Regarding the analysis of output levels, we demonstrated substantial variability of transducers’ characteristics and dynamic range, with headphones yielding higher output levels compared to earphones. These findings confirm the importance of a self-adjusted volume option. These results also suggest that earphones may not be suitable for test execution as the output levels may be relatively low, particularly for subjects with hearing loss or for those who skip the volume adjustment step. Further research is needed to fully address test performance, e.g. testing a larger sample of subjects, addressing different classification approaches, and characterizing test reliability in varying conditions using different devices and transducers.
Lecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering, LNICST
Classification
Decision trees
Hearing loss
Hearing screening
Smartphone app
Speech-in-noise testing
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1168438
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? ND
social impact