@conference{gablenz2018,
 author = {Gablenz, P. v. and Holube, I. and Kowalk, U. and Bilert, S. and Meis, M. and Bitzer, J.},
 title = {Data analysis from real-world hearing assessment},
 abstract = {Ecological Momentary Assessment (EMA) is a promising approach for evaluating the impact of hearing loss and the benefit of rehabilitative interventions in real-world settings. However, largely abandoning controlled test conditions involves challenges of many kinds. It certainly underlines the need for a conceptual and procedural framework that guides both the technical validation and the patient-centered interpretation of the gathered data. In an ongoing EMA field study, a smartphone-based device is used to collect subjective listening assessments by means of an app-based digital questionnaire as well as objective acoustical parameters (see IHCON-contribution Kowalk et al, 2018). Audio signals are received by means of two microphones mounted on the left and right temple of the participant’s glasses. However, the storage of audio features is limited to RMS levels, smoothed Auto and Cross-Power Spectral Densities (A/C-PSD) and Zero-Crossing Rates (ZCR) to ensure privacy. Initially, an ad-hoc analysis is performed when the study participants return the EMA device, basically to inform the supervisor about the hardware runtime, data storage and validity, and the main assessment results sorted by the participants’ activities and locations. Descriptive statistics, e.g. on the frequency of assessments, mean ratings of speech understanding or listening effort are mostly graphically displayed, thus easy to grasp for the supervisor and – complemented by some oral explanations – mostly accessible to the study participant. As this EMA study is primarily designed for rehabilitative purposes, we reject to regard study participants as mere data providers, but try to let them participate as much as possible in their EMA results. In addition, this ad-hoc analysis includes so-called day prints. These figures show RMS level, PSD and assessments versus time, thus carry highly condensed and detailed information useful both to the supervisor and for later data analyses. Subsequently, the real part of the coherence calculated from PSDs is used to identify data segments containing the study participants’ own voice. Whatever approach is applied in order to estimate the level of signals and background noise, reliable detection of own-voice segments is crucial to prevent biased results. Reference on own-voice activity, moreover, allows for cross-validating communication situations reported in the questionnaire. This contribution outlines this two-step data analysis established to cope with EMA data in rehabilitative hearing research and shows exemplary results derived in hearing-impaired persons.},
 tags = {EMA, IHAB-RL, H+A, HALLO},
 booktitle = {International Hearing Aid Research Conference (IHCON), Lake Tahoe, CA, USA},
 year = {2018},
}